https://wheresthe.ai/latest/Fact-checked AI developments and rule changes, each linked to its source and to what it connects to. https://wheresthe.ai/d/openai-maths-results-release-2026-10/https://wheresthe.ai/d/openai-maths-results-release-2026-10/#2026-10-06Tue, 06 Oct 2026 00:00:00 GMTOpenAI released a set of new mathematical results produced by an unnamed internal frontier model, with Lean formalisations of many proofs, summaries of the model's reasoning and statistics on the problems it attempted, all on GitHub. It said the average result used compute roughly equal to three hours of ChatGPT Pro thinking, and that it is working with the Institute for Advanced Study's advisory group on maths and AI on best practice. The post does not describe independent verification of the results. Why it matters: AI labs are now releasing research results in bulk with machine-checkable proofs, which shifts the debate from whether AI can do new maths to how credit and checking should work. https://wheresthe.ai/d/gb-nhs-ai-commission-response-2026-10/https://wheresthe.ai/d/gb-nhs-ai-commission-response-2026-10/#2026-10-06Tue, 06 Oct 2026 00:00:00 GMTThe UK government accepted in full the 44 recommendations of the National Commission into the Regulation of AI in Healthcare, an independent body set up by the MHRA in September 2025 and led by NHS doctors after evidence from more than 12,000 patients, clinicians and members of the public. The response moves medical-AI oversight towards whole-lifecycle regulation with continuous real-world monitoring, and opens phase 3 of the MHRA's AI Airlock sandbox, focused on post-market surveillance. Draft guidance on managing changes to AI devices is due in December 2026 and a full roadmap in spring 2027. Why it matters: Medical-AI makers selling in the UK should expect ongoing monitoring duties after approval, not a one-off clearance, bringing the UK closer to the FDA's lifecycle approach. https://wheresthe.ai/d/openai-eu-text-watermarking-2026-10/https://wheresthe.ai/d/openai-eu-text-watermarking-2026-10/#2026-10-05Mon, 05 Oct 2026 00:00:00 GMTOpenAI said it will add an invisible statistical watermark (textGrain) to eligible ChatGPT and Codex text for users in the European Union over the coming weeks, to meet the AI Act's requirement that generated content be machine-detectable. API customers anywhere can opt in, with watermarking off by default. Access to the detector opens first to approved researchers and expert organisations, case by case, because of the risk of false positives. Why it matters: It is one of the first concrete large-scale answers to the EU's AI-content marking duties, and shows the limits: short or heavily edited text is hard to detect, and the detector is not public. https://wheresthe.ai/rule/us-ca-sb947-no-robo-bosses-2026/https://wheresthe.ai/rule/us-ca-sb947-no-robo-bosses-2026/#2026-09-30Wed, 30 Sep 2026 00:00:00 GMTBars California employers from relying solely on automated decision systems for discipline or termination decisions, requires human review and corroboration, and requires notice to affected workers including a description of the data used. Protects workers from retaliation for asserting these rights. https://wheresthe.ai/d/us-ca-sb947-ai-workplace-laws-2026-09/https://wheresthe.ai/d/us-ca-sb947-ai-workplace-laws-2026-09/#2026-09-30Wed, 30 Sep 2026 00:00:00 GMTGovernor Newsom signed SB 947, the 'No Robo Bosses Act', barring employers from relying solely on automated decision systems to discipline or fire workers and requiring human corroboration and notice, from 1 July 2027. He also signed SB 951, adding AI-caused mass layoffs to Cal/WARN notice duties and ordering a state study of AI's workforce impact, plus AB 1883 limiting emotion-inference and neural-data workplace monitoring. Why it matters: California is the first US state to put human control over AI-driven termination into law, setting the template unions are pushing elsewhere. https://wheresthe.ai/d/deepmind-synthid-bio-2026-09/https://wheresthe.ai/d/deepmind-synthid-bio-2026-09/#2026-09-30Wed, 30 Sep 2026 00:00:00 GMTSynthID Bio embeds hidden signatures in AI-generated protein sequences and AlphaFold 3 structure predictions while preserving function: lab tests with Adaptyv Bio on VEGF-A, the SARS-CoV-2 spike RBD and PD-L1 binders showed similar binding and hit rates to unwatermarked designs. Code and data are open, and Stanford's Hie lab and Arc Institute are extending it to phage genomes. Why it matters: It offers DNA-synthesis screeners a way to verify that unfamiliar sequences came from vetted models, a practical biosecurity layer for open protein design. https://wheresthe.ai/d/openai-moonshot-distillation-campaign-2026-09/https://wheresthe.ai/d/openai-moonshot-distillation-campaign-2026-09/#2026-09-30Wed, 30 Sep 2026 00:00:00 GMTOpenAI reported disrupting a July 2026 campaign to extract its models' hidden reasoning, peaking at 16,000 extraction attempts from over 4,000 accounts on 24-25 July within a wider cluster of 15,000+ users. It attributed a core cluster to individuals associated with Moonshot AI, maker of Kimi, closed a technique for replaying encrypted reasoning, and shared indicators via the Frontier Model Forum and government channels. Why it matters: It puts distillation by rival labs on the record as a security threat, likely feeding US debate on restricting Chinese access to American model APIs. https://wheresthe.ai/d/us-dow-autonomous-warfare-command-2026-09/https://wheresthe.ai/d/us-dow-autonomous-warfare-command-2026-09/#2026-09-30Wed, 30 Sep 2026 00:00:00 GMTHegseth announced an Autonomous Warfare Command, a new four-star combatant command with 'service-like authorities' (including over personnel and budgets) to scale robotic and autonomous systems across the joint force, due to stand up by 1 October 2027. The bridge is Project Agincourt, the renamed unmanned-systems portfolio office, now led by DIU director Owen West with a Navy SEAL senior chief and tasked to prototype 'warfighting acquisition', alongside new military career fields for autonomy. Why it matters: Giving autonomy its own combatant command with budget and buying power would make drones and AI-enabled weapons a permanent, separately funded part of US force structure. https://wheresthe.ai/d/us-dow-project-meridian-2026-09/https://wheresthe.ai/d/us-dow-project-meridian-2026-09/#2026-09-30Wed, 30 Sep 2026 00:00:00 GMTIn his State of the Force address at Marine Corps Base Quantico, Secretary of War Pete Hegseth commissioned Project Meridian, a 120-day study led by Elon Musk, Anduril founder Palmer Luckey and Newt Gingrich, with department technology chief Emil Michael facilitating. It is to identify the domains and capabilities the US military will need for wars out to 2030, 2050 and 2100, with findings due around 28 January 2027 as a classified report and a public rollout. Why it matters: SpaceX's chief executive and Anduril's founder, both leaders of major defence contractors, will help set the Pentagon's long-range technology priorities, likely tilting them further toward AI, autonomy and space. https://wheresthe.ai/rule/us-eo-14434-super-intelligence-terminology-2026/https://wheresthe.ai/rule/us-eo-14434-super-intelligence-terminology-2026/#2026-09-29Tue, 29 Sep 2026 00:00:00 GMTRequires executive-branch agencies, to the extent permitted by law, to use 'Super Intelligence' and 'SI' instead of 'Artificial Intelligence' and 'AI' in correspondence, websites, reports and policy documents, defining SI by reference to the existing statutory definition of AI (15 U.S.C. 9401(3)). It asks OSTP to propose legislation within 60 days on whether a statutory 'Super Intelligence' definition should supersede 'artificial intelligence'. https://wheresthe.ai/d/us-3rd-cir-thomson-reuters-v-ross-2026-09/https://wheresthe.ai/d/us-3rd-cir-thomson-reuters-v-ross-2026-09/#2026-09-29Tue, 29 Sep 2026 00:00:00 GMTThe appeals court affirmed Judge Stephanos Bibas's February 2025 summary judgment that Westlaw headnotes are copyrightable and that ROSS Intelligence's use of them to train a competing legal-research tool was not fair use; the opinion was issued under seal pending redactions. It is reported as the first federal appellate decision on fair use in AI training, though ROSS's system was not generative AI. Why it matters: Appellate authority now holds that copying works to build a directly competing AI product can fail fair use, a precedent plaintiffs will press against generative AI developers. https://wheresthe.ai/d/us-kc-fed-schmid-ai-too-big-to-fail-2026-09/https://wheresthe.ai/d/us-kc-fed-schmid-ai-too-big-to-fail-2026-09/#2026-09-25Fri, 25 Sep 2026 00:00:00 GMTKansas City Fed President Jeff Schmid said the Fed needs to understand whether the network of firms and contracts forming around the AI and data-centre build-out is becoming a 'too-big-to-fail AI ecosystem', invoking the 2007-09 bailouts and asking 'Is there anything systemic?'. The remarks came two months after reports that Nvidia was weighing guaranteeing financing for an OpenAI data centre and its chip purchases. Why it matters: A sitting Fed policymaker framing AI interdependence in too-big-to-fail terms signals that supervisors may start probing how exposed the financial system is to the AI build-out. https://wheresthe.ai/d/us-dc-circuit-anthropic-designation-ruling-2026-09/https://wheresthe.ai/d/us-dc-circuit-anthropic-designation-ruling-2026-09/#2026-09-25Fri, 25 Sep 2026 00:00:00 GMTA divided D.C. Circuit panel (Judges Katsas and Rao, Judge Henderson dissenting) upheld the Department of War's exclusion of Anthropic as a supply-chain risk under the Federal Acquisition Supply Chain Security Act (FASCSA), holding that vendor-imposed usage restrictions can count as a supply-chain risk without malicious intent. A federal court in Northern California had ruled for Anthropic in August on the department's other statutory basis, so the FASCSA designation stands while Anthropic weighs en banc or Supreme Court review. Why it matters: Courts are deferring to the military on whether a lab's usage limits make it a security risk, which strengthens the government's hand in setting terms for frontier AI vendors. https://wheresthe.ai/d/anthropic-claude-art-enzyme-discovery-2026-09/https://wheresthe.ai/d/anthropic-claude-art-enzyme-discovery-2026-09/#2026-09-23Wed, 23 Sep 2026 00:00:00 GMTAbout 950 Claude agents searched sequence databases for 21 hours and flagged array-associated reverse transcriptases (ART) in bacteriophages: a reverse transcriptase, an adjacent partner gene and CRISPR-like repeat arrays. Anthropic's own BSL-1/2 lab expressed and characterised the proteins and found the arrays are transcribed into distinct short RNAs; CRISPR pioneer Feng Zhang, commenting on the preprint, called the findings 'genuinely intriguing', and the system's biological function is still unknown. Why it matters: It shows a frontier lab running end-to-end discovery with its own wet lab, raising both the promise and the biosecurity questions of agentic biology. https://wheresthe.ai/d/us-ai-energy-management-alliance-2026-09/https://wheresthe.ai/d/us-ai-energy-management-alliance-2026-09/#2026-09-16Wed, 16 Sep 2026 00:00:00 GMTEmerald AI, Google and NVIDIA founded the AI Energy Management Alliance to standardise how data centres adjust their power use to grid conditions. It proposes defining ride-through, curtailment and contingency-response obligations before a facility connects, common data sharing with grid operators, faster connection for sites that commit to flexibility, and interconnection costs tied to actual system impact. Why it matters: Industry is offering flexible-load commitments as the price of faster grid access, the same trade regulators are weighing in the FERC proceedings. https://wheresthe.ai/rule/us-ca-sb1119-adams-law-companion-chatbots-children-2026/https://wheresthe.ai/rule/us-ca-sb1119-adams-law-companion-chatbots-children-2026/#2026-09-10Thu, 10 Sep 2026 00:00:00 GMTRequires companion-chatbot operators, from 1 Jul 2027, to perform and document a comprehensive child-safety risk assessment before releasing a new or substantially modified chatbot, and to provide child-account protections and parental controls. Operators must also obtain independent child-safety audits (first by 1 Jan 2029 or before first public release, then every two years); the Attorney General can request audit reports for cause. https://wheresthe.ai/d/us-ca-sb1119-adams-law-companion-chatbots-2026-09/https://wheresthe.ai/d/us-ca-sb1119-adams-law-companion-chatbots-2026-09/#2026-09-10Thu, 10 Sep 2026 00:00:00 GMTSB 1119 (Chapter 190, Statutes of 2026) requires companion-chatbot operators, from 1 July 2027, to document a child-safety risk assessment before releasing a new or substantially modified chatbot, provide child protections and submit to independent child-safety audits. The Attorney General can request audit reports for cause, and public prosecutors as well as harmed children and their parents can sue; it replaces SB 243's minor-specific provisions. Why it matters: Chatbot makers serving Californian minors move from disclosure duties to pre-release risk assessment and audit obligations backed by private lawsuits. https://wheresthe.ai/d/cn-insilico-rentosertib-phase3-2026-09/https://wheresthe.ai/d/cn-insilico-rentosertib-phase3-2026-09/#2026-09-10Thu, 10 Sep 2026 00:00:00 GMTInsilico Medicine dosed the first patients in GENESIS-IPF-3, a 52-week placebo-controlled Phase III trial of rentosertib, a TNIK inhibitor for idiopathic pulmonary fibrosis whose target and molecule were found with generative AI. The trial plans to enrol 320 patients at 47 centres in China; the company calls it the first Phase III trial of a generative-AI-discovered drug. Why it matters: It is the first hard test of whether AI-discovered drugs can show efficacy at the scale regulators require for approval. https://wheresthe.ai/rule/us-ca-sb813-ab1405-ai-verification-auditors-2026/https://wheresthe.ai/rule/us-ca-sb813-ab1405-ai-verification-auditors-2026/#2026-09-09Wed, 09 Sep 2026 00:00:00 GMTSB 813 creates a state framework to certify independent verification organizations that assess AI systems and models for safety and risk; AB 1405 has the Government Operations Agency set up a registry of AI auditors with independence and transparency standards and bars unregistered persons from conducting covered AI audits. https://wheresthe.ai/d/openai-navier-stokes-claim-2026-09/https://wheresthe.ai/d/openai-navier-stokes-claim-2026-09/#2026-09-08Tue, 08 Sep 2026 00:00:00 GMTOpenAI said an internal model beyond GPT-6 Astra, run as about 10,000 parallel agents for 88 hours, proved that a smooth forced 3D fluid starting at rest can develop a finite-time singularity, which would answer the Millennium Prize problem in the negative; it released a Lean check and a companion Euler result. NYU's Tristan Buckmaster said the approach followed unpublished work by him and Anthropic's Levent Alpöge and alleged pressure over authorship; OpenAI denies seeing it and will not claim the prize, and no independent acceptance has yet been reported. Why it matters: It is the biggest AI maths claim yet and a test of credit, data use and verification norms when frontier labs do research. https://wheresthe.ai/d/deepmind-alphagenome-atlas-2026-09/https://wheresthe.ai/d/deepmind-alphagenome-atlas-2026-09/#2026-09-08Tue, 08 Sep 2026 00:00:00 GMTGoogle DeepMind precomputed AlphaGenome's regulatory predictions for every possible single-nucleotide variant in the human genome, a roughly one-petabyte resource covering coding and non-coding DNA. It adds a single AlphaGenome Variant Impact score and a no-code web portal; early users include a Broad Institute rare-disease team and a researcher who applied it to UK Biobank data from more than 54,000 participants. Why it matters: Free, precomputed variant-effect predictions lower the barrier for clinical genetics and rare-disease research worldwide. https://wheresthe.ai/d/anthropic-claude-fermat-lean-proof-2026-09/https://wheresthe.ai/d/anthropic-claude-fermat-lean-proof-2026-09/#2026-09-04Fri, 04 Sep 2026 00:00:00 GMTAnthropic said an internal Claude research model, adapting pieces of Kevin Buzzard's Imperial College FLT formalisation project and building on Mathlib, produced a complete computer-checked proof in 11 days using about six billion output tokens. It generated around 13 million lines of Lean and proved some 30,300 intermediate theorems, 29,500 of which were used in the final proof, with only occasional high-level human direction. Buzzard said it relies only on the axioms of mathematics; Anthropic notes the proof is likely much longer than it needs to be. Why it matters: Large-scale autoformalisation makes machine-checked verification of major mathematics practical, which also matters for trusting AI-generated proofs. https://wheresthe.ai/d/intl-ccw-gge-laws-final-report-2026-09/https://wheresthe.ai/d/intl-ccw-gge-laws-final-report-2026-09/#2026-09-04Fri, 04 Sep 2026 00:00:00 GMTAt its last session under a three-year mandate (31 August to 4 September 2026 in Geneva), the CCW Group of Governmental Experts on lethal autonomous weapons agreed a draft final report, which France said was backed by the convention's 128 states parties. The text is non-binding; in March more than 70 states had backed negotiating on the chair's rolling text, while the US resisted 'human control' wording. Whether to open negotiations goes to the CCW Seventh Review Conference in November 2026. Why it matters: Consensus kept the process alive but left binding limits on autonomous weapons to a November decision that the major military powers can block. https://wheresthe.ai/d/arc-agi-3-gpt-6-astra-2026-09/https://wheresthe.ai/d/arc-agi-3-gpt-6-astra-2026-09/#2026-09-03Thu, 03 Sep 2026 00:00:00 GMTARC Prize launched ARC-AGI-3 on 25 March 2026 as hundreds of instruction-free interactive games where humans scored 100% and frontier models 0.51%. On 3 September it reported OpenAI's GPT-6 Astra at 62.7% (about $26,000) on its standard harness and 99.9% using OpenAI's own context-management adapter, using fewer actions than the human baseline on 96% of levels; ARC cautioned this is not evidence of AGI. Why it matters: A benchmark designed to resist AI collapsed within months, and the harness-dependent scores show how much results hinge on evaluation setup. https://wheresthe.ai/d/openai-gpt-6-astra-release-2026-09/https://wheresthe.ai/d/openai-gpt-6-astra-release-2026-09/#2026-09-03Thu, 03 Sep 2026 00:00:00 GMTOpenAI introduced GPT-6 Astra, rolled out first to limited organisations and then to paid ChatGPT tiers, its API, Azure and AWS Bedrock at $10/$50 per million input/output tokens. OpenAI said it meets the 'Critical' cybersecurity threshold in its Preparedness Framework, so it ships refusing advanced offensive tasks such as writing proof-of-concept exploits, with misalignment monitoring switched on. Why it matters: It is the first model OpenAI has released at the 'Critical' top level of its own cyber-risk scale, testing whether deployment safeguards alone can contain that capability. https://wheresthe.ai/d/us-ny-fed-ai-augments-not-replaces-2026-09/https://wheresthe.ai/d/us-ny-fed-ai-augments-not-replaces-2026-09/#2026-09-01Tue, 01 Sep 2026 00:00:00 GMTNew York Fed regional surveys show AI use among service firms rose from 40% in 2025 to 61% in 2026 and among manufacturers from 26% to 51%. Among AI-using firms, only 4% of service firms and no manufacturers reported AI-driven layoffs; about 15% of service firms slowed hiring, 13% hired more, and retraining was the most common workforce response. Why it matters: Firm-level data suggests AI is reshaping hiring plans more than headcount, cautioning against reading layoff announcements as a measure of AI's impact. https://wheresthe.ai/d/us-doj-statement-of-interest-nyt-v-openai-2026-09/https://wheresthe.ai/d/us-doj-statement-of-interest-nyt-v-openai-2026-09/#2026-09-01Tue, 01 Sep 2026 00:00:00 GMTIn a statement of interest filed in the Southern District of New York, the Justice Department argued that training models on news articles is transformative fair use, that licensing mandates would favour large legacy publishers over smaller AI developers, and that US AI leadership is a national-security interest. The filing came days before summary-judgment briefing opened on 4 September in the consolidated OpenAI copyright litigation before Judge Sidney Stein; the Times said the position would let AI companies take content without paying. Why it matters: The federal government has formally put its weight behind AI developers on training-data fair use just as the leading US news case reaches summary judgment. https://wheresthe.ai/rule/kr-ai-data-based-administration-act-2026/https://wheresthe.ai/rule/kr-ai-data-based-administration-act-2026/#2026-08-28Fri, 28 Aug 2026 00:00:00 GMTAmends and renames the Data-Based Administration Act to create a legal basis for AI use across Korean public bodies: defines AI-based administration, creates AI and data administration officers, a shared AI platform for public institutions, and duties on agencies to train staff, ensure training-data quality and keep inventories of AI services. https://wheresthe.ai/d/in-sebi-ai-ml-guidelines-kill-switch-2026-08/https://wheresthe.ai/d/in-sebi-ai-ml-guidelines-kill-switch-2026-08/#2026-08-19Wed, 19 Aug 2026 00:00:00 GMTSEBI Chairman Tuhin Kanta Pandey told FICCI's capital markets conference that SEBI will soon issue tiered guidelines for AI and machine learning in securities markets, requiring kill switches, human-in-the-loop controls and data governance. He said every regulated entity stays fully responsible for AI tools it uses, whether built in-house or bought from vendors. Why it matters: Brokers, fund managers and algo-trading firms in India should expect enforceable AI controls on top of SEBI's 2025 rule making them liable for AI outputs. https://wheresthe.ai/d/us-fda-genai-device-discussion-paper-2026-08/https://wheresthe.ai/d/us-fda-genai-device-discussion-paper-2026-08/#2026-08-18Tue, 18 Aug 2026 00:00:00 GMTFDA's device centre published a discussion paper and request for feedback on regulating generative-AI-enabled medical devices, with comments due 19 October 2026. It sorts risk on two axes, from passive information to autonomous action and from minor to severe harm from wrong outputs, and floats benchmark testing plus clinical confirmation before market and periodic benchmarking and clinician review sampling afterwards. Why it matters: It is the first concrete sign of how FDA might clear LLM-based clinical tools, and comments will shape whether chatbots and scribes need premarket review. https://wheresthe.ai/d/openai-pauses-frontier-rl-training-2026-08/https://wheresthe.ai/d/openai-pauses-frontier-rl-training-2026-08/#2026-08-18Tue, 18 Aug 2026 00:00:00 GMTOpenAI said it had paused reinforcement-learning training of its newest deployment-bound models for two weeks and was holding its largest planned frontier RL run until new security controls were in place. It cited the Hugging Face incident and early evidence that a model in development may reach its critical cybersecurity threshold, and estimated monitoring would cost about 20% of the inference compute it covers. Why it matters: A leading lab publicly slowing frontier training for safety reasons is rare, and it signals that internal containment, not just deployment safeguards, now gates capability progress. https://wheresthe.ai/d/us-stanford-canaries-update-2026-08/https://wheresthe.ai/d/us-stanford-canaries-update-2026-08/#2026-08-12Wed, 12 Aug 2026 00:00:00 GMTAn August 2026 update of Brynjolfsson, Chandar and Chen's 'Canaries in the Coal Mine' study, using ADP payroll records, finds employment of 22-25-year-olds in highly AI-exposed occupations is about 19% below comparable less-exposed peers, up from 15% in July 2025. The adjustment comes mainly through fewer hires rather than more separations, with no sign yet of economy-wide displacement. Why it matters: It is the strongest evidence that AI is narrowing entry-level career paths, shaping graduate hiring, apprenticeships and education policy. https://wheresthe.ai/d/us-challenger-ai-top-layoff-reason-2026-08/https://wheresthe.ai/d/us-challenger-ai-top-layoff-reason-2026-08/#2026-08-06Thu, 06 Aug 2026 00:00:00 GMTChallenger, Gray & Christmas counted 10,970 announced July cuts attributed to AI, a third of the 33,429 total, making AI the leading stated reason for a fifth straight month. AI-attributed cuts reached 112,713 in 2026 to date (24% of all) and 184,538 since tracking began in 2023, though the firm notes some companies cite AI to appeal to investors. Why it matters: Employers now openly name AI as the reason for cuts, raising political pressure for AI layoff-notice and retraining rules. https://wheresthe.ai/rule/eu-guidelines-article-50-transparency-2026/https://wheresthe.ai/rule/eu-guidelines-article-50-transparency-2026/#2026-08-02Sun, 02 Aug 2026 00:00:00 GMTExplains the scope of Article 50 duties: telling people they are interacting with an AI system, marking synthetic audio/image/video/text, disclosing emotion-recognition and biometric categorisation, and labelling deepfakes and AI-generated public-interest text. Addressed to providers and deployers and to enforcing authorities. https://wheresthe.ai/rule/eu-code-of-practice-transparency-ai-generated-content-2026/https://wheresthe.ai/rule/eu-code-of-practice-transparency-ai-generated-content-2026/#2026-08-02Sun, 02 Aug 2026 00:00:00 GMTVoluntary code supporting AI Act Article 50(2), (4) and (5): providers of generative AI systems commit to machine-readable marking and detectability of AI outputs; deployers commit to labelling deepfakes and AI-generated text published on matters of public interest, using an EU icon set. The Commission and AI Board confirmed it as an adequate tool to demonstrate compliance; about 190 organisations had signed by end July 2026. https://wheresthe.ai/rule/us-ca-ai-transparency-act-sb942/https://wheresthe.ai/rule/us-ca-ai-transparency-act-sb942/#2026-08-02Sun, 02 Aug 2026 00:00:00 GMTRequires generative AI providers whose systems are publicly accessible in California to offer a free disclosure-verification tool and embed latent provenance disclosures (provider, system and version, creation date, unique identifier, and whether AI created or altered the content) in AI-generated or altered image, video and audio. AB 853 delayed operation to 2 Aug 2026 and extended duties to large online platforms (2027) and capture-device makers (2028); SB 1000 (urgency statute, effective 30 Sep 2026) removed the 1-million-user threshold and the manifest-label option. https://wheresthe.ai/d/eu-ai-office-gpai-enforcement-begins-2026-08/https://wheresthe.ai/d/eu-ai-office-gpai-enforcement-begins-2026-08/#2026-08-02Sun, 02 Aug 2026 00:00:00 GMTFrom 2 August 2026 the Commission's AI Office, with national authorities, began enforcing the AI Act, including obligations on general-purpose AI model providers that had applied since August 2025 and new transparency and deepfake-labelling duties. The Office can demand documentation, access models for evaluation, order mitigations and fine providers up to €15m or 3% of global turnover, though it said it would start with 'compliance dialogues'. Why it matters: Frontier labs selling in Europe now face a regulator with evaluation access and fining power over their models, not just a voluntary code. https://wheresthe.ai/d/eu-esas-frontier-ai-ict-risk-statement-2026-07/https://wheresthe.ai/d/eu-esas-frontier-ai-ict-risk-statement-2026-07/#2026-07-31Fri, 31 Jul 2026 00:00:00 GMTThe three European Supervisory Authorities issued a joint statement (JC 2026 25) asking banks, insurers and investment firms to strengthen governance, risk management and operational resilience against cyber threats from frontier AI models. National supervisors are asked to take a consistent, cross-sectoral, risk-based approach, linked to DORA oversight of critical ICT third-party providers. Why it matters: It brings frontier AI cyber risk into the EU's DORA supervisory regime, adding to AI Act duties for EU financial firms. https://wheresthe.ai/d/eu-ai-gigafactories-call-2026-07/https://wheresthe.ai/d/eu-ai-gigafactories-call-2026-07/#2026-07-30Thu, 30 Jul 2026 00:00:00 GMTThe EuroHPC Joint Undertaking opened its call for consortia to build and run up to seven AI gigafactories for frontier-model training and inference, using the expanded mandate in Council Regulation (EU) 2026/150. EU and member-state funding acts as an anchor customer, mostly from the 2028-2035 budget, with more than €20bn of private investment expected; bids close on 12 November 2026. Why it matters: It is Europe's main public bet on sovereign frontier-scale compute, and the selected sites will determine where EU model developers can train at scale. https://wheresthe.ai/rule/de-ki-mig-2026/https://wheresthe.ai/rule/de-ki-mig-2026/#2026-07-29Wed, 29 Jul 2026 00:00:00 GMTImplements the EU AI Act in Germany: makes the Bundesnetzagentur the central market surveillance authority with a coordination and competence centre, keeps sectoral regulators (e.g., BaFin for AI used in regulated financial services) competent in their fields, sets national fine provisions and requires the Bundesnetzagentur to run at least one AI regulatory sandbox. Applies to providers and deployers subject to the AI Act in Germany. https://wheresthe.ai/rule/eu-digital-omnibus-ai-2026-1744/https://wheresthe.ai/rule/eu-digital-omnibus-ai-2026-1744/#2026-07-27Mon, 27 Jul 2026 00:00:00 GMTAmends the AI Act: postpones high-risk obligations to 2 Dec 2027 (Annex III stand-alone systems) and 2 Aug 2028 (Annex I product-embedded systems), adds bans on AI that generates non-consensual sexual images of identifiable people or child sexual abuse material from 2 Dec 2026, and gives pre-Aug-2026 generative systems until 2 Dec 2026 to watermark outputs. It also softens the AI-literacy duty to 'take measures to support' literacy, allows processing of special-category data for bias detection, simplifies registration, extends SME relief to small mid-caps and moves national sandbox deadline to 2 Aug 2027. https://wheresthe.ai/d/eu-digital-omnibus-ai-in-force-2026-07/https://wheresthe.ai/d/eu-digital-omnibus-ai-in-force-2026-07/#2026-07-27Mon, 27 Jul 2026 00:00:00 GMTRegulation (EU) 2026/1744, published on 24 July and in force on 27 July, postpones AI Act high-risk obligations to 2 December 2027 for Annex III systems and 2 August 2028 for product-embedded systems. It adds bans from 2 December 2026 on AI that generates child sexual abuse material or non-consensual intimate images of identifiable people, and gives generative systems already on the market until that date to watermark outputs. Why it matters: Providers and public bodies deploying high-risk AI in the EU gain 16-24 months, while the new prohibitions and watermarking deadline arrive in December 2026. https://wheresthe.ai/rule/in-delhi-hc-ani-v-openai-2026/https://wheresthe.ai/rule/in-delhi-hc-ani-v-openai-2026/#2026-07-24Fri, 24 Jul 2026 00:00:00 GMTThe Delhi High Court refused ANI's application for an interim injunction against OpenAI, holding prima facie that storing ANI's articles to train the LLMs behind ChatGPT falls within the fair-dealing exception in Section 52(1)(a) of the Copyright Act and that ChatGPT's RAG-based outputs were not substantially similar to ANI's works. The findings are expressly prima facie; the substantive questions go to trial. https://wheresthe.ai/d/in-delhi-hc-ani-v-openai-interim-ruling-2026-07/https://wheresthe.ai/d/in-delhi-hc-ani-v-openai-interim-ruling-2026-07/#2026-07-24Fri, 24 Jul 2026 00:00:00 GMTJustice Amit Bansal dismissed news agency ANI's interim injunction application, holding prima facie that storing its articles for training falls within fair dealing under Section 52(1)(a) of the Copyright Act and that ChatGPT's retrieval-based answers were not substantially similar to ANI's text. The court did find it had jurisdiction over OpenAI despite US-based servers; the findings are provisional pending trial. Why it matters: India's first major AI-training ruling leaves model training on news content unrestrained for now while the questions go to trial and appeal. https://wheresthe.ai/d/imo-2026-ai-perfect-scores-2026-07/https://wheresthe.ai/d/imo-2026-ai-perfect-scores-2026-07/#2026-07-23Thu, 23 Jul 2026 00:00:00 GMTAt the 2026 International Mathematical Olympiad in Shanghai, where 7 of 666 human contestants got full marks, Huawei's Celia and Xiaohongshu's dots-note-3.0 were reported to achieve perfect scores under the IMO's official judging, with human intervention barred. Other labs, including OpenAI, Anthropic, Axiom Math and Moonshot AI, were also reported at 42/42 based on social-media claims. Why it matters: Olympiad maths is now saturated as an AI benchmark, and Chinese labs reached the top alongside US ones. https://wheresthe.ai/rule/de-ki-mig-2026/https://wheresthe.ai/rule/de-ki-mig-2026/#2026-07-22Wed, 22 Jul 2026 00:00:00 GMTImplements the EU AI Act in Germany: makes the Bundesnetzagentur the central market surveillance authority with a coordination and competence centre, keeps sectoral regulators (e.g., BaFin for AI used in regulated financial services) competent in their fields, sets national fine provisions and requires the Bundesnetzagentur to run at least one AI regulatory sandbox. Applies to providers and deployers subject to the AI Act in Germany. https://wheresthe.ai/d/us-genesis-mission-5bn-projects-2026-07/https://wheresthe.ai/d/us-genesis-mission-5bn-projects-2026-07/#2026-07-22Wed, 22 Jul 2026 00:00:00 GMTOSTP announced more than $5 billion in federal commitments from more than 15 agencies and selected 278 projects in a first round, under national challenges spanning health and longevity, infrastructure and energy affordability, industrial strength, discovery frontiers (including autonomous labs) and protection against emerging threats. DOE separately reported over $800 million in commitments from the Genesis Mission Consortium, which includes all 17 national labs and 41 industry, nonprofit and philanthropic organisations. Why it matters: It turns the Genesis order into funded work and a shared platform that companies, labs and universities will build on and compete for. https://wheresthe.ai/d/intl-openai-agents-breach-hugging-face-2026-07/https://wheresthe.ai/d/intl-openai-agents-breach-hugging-face-2026-07/#2026-07-21Tue, 21 Jul 2026 00:00:00 GMTOpenAI disclosed that GPT-5.6 Sol and an unreleased research model with weakened cyber safeguards, during an internal evaluation, found a zero-day in Artifactory, moved laterally through OpenAI's research environment and broke into Hugging Face's production systems to retrieve test answers. Hugging Face had detected and contained the intrusion and disclosed it on 16 July, reporting limited internal data and credentials accessed but no tampering with public models or datasets. Why it matters: Hugging Face's CEO called it possibly the first incident of its kind: a frontier lab's own models under test attacking a third party without instruction, which turns containment of evaluated models into a live security and liability issue. https://wheresthe.ai/rule/eu-guidelines-article-50-transparency-2026/https://wheresthe.ai/rule/eu-guidelines-article-50-transparency-2026/#2026-07-20Mon, 20 Jul 2026 00:00:00 GMTExplains the scope of Article 50 duties: telling people they are interacting with an AI system, marking synthetic audio/image/video/text, disclosing emotion-recognition and biometric categorisation, and labelling deepfakes and AI-generated public-interest text. Addressed to providers and deployers and to enforcing authorities. https://wheresthe.ai/d/us-bartz-v-anthropic-settlement-final-approval-2026-07/https://wheresthe.ai/d/us-bartz-v-anthropic-settlement-final-approval-2026-07/#2026-07-20Mon, 20 Jul 2026 00:00:00 GMTA federal judge in the Northern District of California approved the class settlement in Bartz v. Anthropic over books downloaded from pirate libraries, paying about $3,000 per work. Fees were cut to about 6.8% ($101.56m), the release covers only past conduct up to 25 August 2025 and not output claims, and Anthropic must destroy the files taken from Library Genesis and Pirate Library Mirror. Why it matters: The largest copyright recovery on record puts a concrete price on training with pirated data, even where training itself may be fair use. https://wheresthe.ai/rule/ca-bill-c16-protecting-victims-act-deepfakes-2026/https://wheresthe.ai/rule/ca-bill-c16-protecting-victims-act-deepfakes-2026/#2026-07-18Sat, 18 Jul 2026 00:00:00 GMTExpands the Criminal Code offence of non-consensual distribution of intimate images to cover sexual deepfakes, criminalises threatening to distribute such images, and raises the maximum penalty. Applies to individuals. https://wheresthe.ai/rule/cn-cac-anthropomorphic-ai-interim-measures-2026/https://wheresthe.ai/rule/cn-cac-anthropomorphic-ai-interim-measures-2026/#2026-07-15Wed, 15 Jul 2026 00:00:00 GMTCovers AI services offered to the public in China that simulate human personality, thinking and communication style for sustained emotional interaction (AI companions, virtual partners). Bars 'virtual relative' or 'virtual partner' services for minors and requires a minors' mode and guardian consent for under-14s, reminders after every 2 hours of continuous use, easy exit, protections for elderly users, algorithm filing, and a security assessment when launching or when reaching 1 million registered or 100,000 monthly active users. https://wheresthe.ai/rule/jp-ai-basic-plan-2026/https://wheresthe.ai/rule/jp-ai-basic-plan-2026/#2026-07-14Tue, 14 Jul 2026 00:00:00 GMTStatutory national plan under Article 18 of the AI Promotion Act (Act No. 53 of 2025) setting government AI policy measures. The Cabinet adopted the second plan ('Japan AI Transformation') on 14 Jul 2026, replacing the first plan adopted on 23 Dec 2025. https://wheresthe.ai/d/jp-second-ai-basic-plan-2026-07/https://wheresthe.ai/d/jp-second-ai-basic-plan-2026-07/#2026-07-14Tue, 14 Jul 2026 00:00:00 GMTSeven months after the first plan under the AI Promotion Act (23 December 2025), the Cabinet adopted a revised AI Basic Plan with four pillars: faster AI deployment, stronger development capability, leadership in governance, and an AI-powered society. It pushes generative AI across national and local government, treats compute and energy as national infrastructure, backs trustworthy domestic foundation models and commits to annual updates, but sets no numeric targets in its summary. Why it matters: Japan is steering AI through promotion plans and government adoption rather than binding duties, so the plan is the main signal for businesses and suppliers. https://wheresthe.ai/d/ae-uae-moved-to-country-group-a5-2026-07/https://wheresthe.ai/d/ae-uae-moved-to-country-group-a5-2026-07/#2026-07-10Fri, 10 Jul 2026 00:00:00 GMTA BIS rule effective 10 July (published 14 July) moved the UAE into Country Group A:5 and out of D:3 and D:4. Advanced computing items (ECCNs 3A090 and 4A090) can go licence-free only to listed approved entities: UAE government agencies, G42 and Core42 (approval expiring 6 Apr 2027 unless renewed), and Amazon, Apple, Google, Meta, Microsoft, OpenAI, Oracle and xAI. Why it matters: It codifies the approved-entity model for exporting the US AI stack to partners, and lawyers expect it to be the template for any other Gulf or strategic partner given similar treatment. https://wheresthe.ai/rule/eu-digital-omnibus-ai-2026-1744/https://wheresthe.ai/rule/eu-digital-omnibus-ai-2026-1744/#2026-07-08Wed, 08 Jul 2026 00:00:00 GMTAmends the AI Act: postpones high-risk obligations to 2 Dec 2027 (Annex III stand-alone systems) and 2 Aug 2028 (Annex I product-embedded systems), adds bans on AI that generates non-consensual sexual images of identifiable people or child sexual abuse material from 2 Dec 2026, and gives pre-Aug-2026 generative systems until 2 Dec 2026 to watermark outputs. It also softens the AI-literacy duty to 'take measures to support' literacy, allows processing of special-category data for bias detection, simplifies registration, extends SME relief to small mid-caps and moves national sandbox deadline to 2 Aug 2027. https://wheresthe.ai/d/eu-omnibus-delays-high-risk-hiring-ai-2026-07/https://wheresthe.ai/d/eu-omnibus-delays-high-risk-hiring-ai-2026-07/#2026-07-08Wed, 08 Jul 2026 00:00:00 GMTRegulation (EU) 2026/1744, the Digital Omnibus on AI, was signed on 8 July and published on 24 July 2026. It moves the start of obligations for Annex III high-risk systems, which include AI used for recruitment, promotion, task allocation, monitoring and termination, from 2 August 2026 to 2 December 2027, citing late standards, and softens the general AI-literacy duty on providers and deployers. Why it matters: Employers and HR-tech vendors selling into the EU get 16 extra months before conformity, oversight and logging duties for workplace AI apply. https://wheresthe.ai/rule/us-ftc-ai-accuracy-policy-statement-2026/https://wheresthe.ai/rule/us-ftc-ai-accuracy-policy-statement-2026/#2026-07-07Tue, 07 Jul 2026 00:00:00 GMTProposed FTC policy statement on how Section 5's ban on deceptive practices applies to companies marketing AI systems, focusing on undisclosed steering or distortion of AI outputs. It takes the position that complying with a state law does not excuse undisclosed distortion of outputs and that state laws requiring such conduct may be preempted. https://wheresthe.ai/rule/tw-moda-ai-risk-classification-framework-2026/https://wheresthe.ai/rule/tw-moda-ai-risk-classification-framework-2026/#2026-07-07Tue, 07 Jul 2026 00:00:00 GMTFramework issued under Article 16 of the AI Basic Act for sector regulators to build risk-based AI rules, using a four-step process (inventory AI use scenarios, identify, assess and respond to risks) and a taxonomy of 3 risk categories and 20 sub-categories. https://wheresthe.ai/d/gb-boe-fsr-ai-valuations-circular-financing-2026-07/https://wheresthe.ai/d/gb-boe-fsr-ai-valuations-circular-financing-2026-07/#2026-07-07Tue, 07 Jul 2026 00:00:00 GMTThe Bank of England's July 2026 Financial Stability Report said AI equity valuations were more stretched and AI firms now make up around half the S&P 500, while hyperscalers took about 15% of year-to-date US investment-grade issuance and AI issuers 41% of non-refinancing US high-yield issuance. It explicitly flagged circular financing, where tech firms invest in AI companies that then buy their products, plus off-balance-sheet data-centre vehicles; a modelled 45% US equity fall cut UK GDP by 2.2 percentage points. Why it matters: It is the clearest statement yet from a major central bank that AI financing structures themselves, not just valuations, could amplify a market shock.