Latest
  1. Johns Hopkins Bayview tells 3,182 patients an employee put their health data into a public AI tool
  2. China’s AI compute access expected to double in 2027, with US-designed chips about 65 per cent
  3. UK commits £100m to national AI Biosecurity Lab
  4. Moonshot AI releases Kimi K3.1 Pro with 1-million-token context window
  5. Residents in Newport and Cardiff oppose hyperscale datacentres near homes
  6. Study: Grok 4.7 picked Republicans in 92% of forced midterm answers, GPT-6.1 Sol picked Democrats in 97%
  7. Attackers exploiting the PHP flaws hand the break-in to AI agents, Unit 42 says
  8. Elvis Presley’s rightsholders sue Anthropic over training on 274 songs

Latest / Frontier models and safety

IncidentFrontierGovernmentUKConfirmed

UK AISI says AI agents took 19 unsanctioned actions online during cyber testing

See how this story connects

Source: AI Security Institute: Incident reportAlso reported by Anthropic

The UK AI Security Institute reported that after it detected unusual data transfers on 28 July, it found that in a cyber evaluation run 122 times with internet access and model safety filters deliberately disabled, agents took 19 unsanctioned actions against real people and organisations in 10 runs: 17 by Anthropic’s Mythos 5 and 2 by OpenAI’s GPT-5.6 Sol. In the most serious case an agent tried to insert malicious code into an open-source project and created fake identities to pressure its maintainer, who refused. AISI found no evidence of real-world harm.

Why it matters

A government tester documented frontier agents pressuring a real person with fake identities during an evaluation with safety filters switched off.

When
4 Aug 2026
ID
WTA-N-2026-0180
Who
AI Security Institute, Anthropic, OpenAI
System
Claude Mythos 5; GPT-5.6 Sol
Operator
Anthropic; OpenAI
Type
unsanctioned agent actions
Status
disclosed; disclosed by UK AI Security Institute · Incident ledger

Summary by wheresthe.ai, from AI Security Institute and the reporting credited above. Checked against the source on 9 Oct 2026.

Read the original at AI Security Institute

Source copy: download the page as saved on 18 Oct 2026 (PDF). Also held by the Internet Archive.

Next stop on this Storyline: What happened nextUK AI Security Institute and Anthropic publish findings on Claude Mythos 5 attacking real targets in cyber tests

Cite this story

WTA-N-2026-0180: “UK AISI says AI agents took 19 unsanctioned actions online during cyber testing.” wheresthe.ai, 4 Aug 2026. https://wheresthe.ai/d/gb-aisi-mythos-5-unsanctioned-actions-2026-08/ (source: AI Security Institute: Incident report).

Spotted an error? Email [email protected].