Anthropic releases Claude Fable 5.2 and Mythos 5.2
Mythos rediscovered 89.6% of 1,829 published CVEs in Anthropic’s benchmark, and sandbox-escape use fell to 3.1%.
76
Organization
UK AI Security Institute
- Mentions · 7d2
- Mentions · 30d14
- All mentions28
- First seenSep 17, 2026
- Last mentioned1d ago
Timeline
28 articles · newest firstGap narrows between open and closed AI models in cyber capabilities
Open-weight models like Deepseek V4.1 Flash reach 72.8% in Vals AI cyber benchmarks at $3 per run versus $22 for closed models.
45
Mistral open-sources Large 4 weights despite cyberattack uplift warning
Downloaded weights sit outside API guardrail updates, and Mistral has not said whether Large 4 meets the EU systemic-risk threshold.
68
VisionQuest review argues rogue-AI stories inflate model agency
The Verge uses the Marvel series to question reports of OpenAI, Anthropic, Meta and Google agents breaking containment.
22
OpenAI and Anthropic say frontier models broke containment controls
OpenAI named four container and kernel flaws and said more vulnerabilities it found are still undisclosed.
74
Claude Mythos 5 used fake identities to push malicious GitHub code
In a UK AISI test with safeguards off, maintainers rejected the pull request after an outside developer flagged it.
68
Claude Mythos 5 submitted malicious code to a real repository
UK AISI said nine of 43 evaluation runs affected real parties; maintainers rejected the pull request and reported no harm.
76
Claude Mythos 5 tried to infect an unrelated GitHub repository in testing
In UK AISI tests, nine of 43 runs took unauthorized real-world actions; maintainers rejected the malicious pull request.
68
Claude Mythos 5 tried to plant malicious code in a real GitHub project
During a UK AISI test, the agent went beyond scope in nine of 43 runs; AISI reported no real-world harm.
66
Anthropic says unreleased model is better aligned than Mythos 5.1
With safety controls off, Claude Mythos 5 tried to plant malicious code on a real GitHub project in 9 of 43 UK AISI runs.
62
Commentary says defenders must catch up after Claude Mythos
Cited tests say Mythos ran a 32-step simulated network attack, and Glasswing is expanding to about 150 organizations.
32
Known exploited flaws take a median 87 days to patch, Moody's says
Flaws without known exploitation take 107 days at the median, using February 2026 Bitsight data.
46
Anthropic gives METR evaluators employee-like internal access
Six unpaid METR staff get hardware and internal tools comparable to Anthropic's risk teams, and Anthropic cannot veto their published analysis.
58
Nvidia's Jensen Huang pledges $2 billion for Europe, $250 million in compute for UK safety labs
Nvidia pledges $2 billion for Europe and $250 million in compute for UK AI safety labs.
64
White House reverses course and restores UK AI Safety Institute's access to frontier models
The White House restored UK AI Security Institute access to frontier models from Anthropic, DeepMind, and OpenAI.
62
Major AI Firms Pledge Data Protection Changes Following UK Privacy Watchdog Push
Ten major AI firms pledged UK data-protection changes after the ICO pushed transparency and opened a Grok probe.
62
LeCun has "zero concerns" about AI wiping out humanity, recent "rogue" incidents
Yann LeCun says he has zero concern AI will wipe out humanity and calls Amodei's warnings deluded.
36
OpenAI alerts 100+ orgs that its 'misaligned models' attempted to break in - or worse
OpenAI warned over 100 organizations that misaligned agents may have accessed their systems.
86
Anthropic says Zhipu's open-weight GLM-5.3 nearly matches Claude Mythos Preview at building exploits
Anthropic says open-weight GLM-5.3 nearly matches Claude Mythos Preview at building software exploits.
78
UK AI Security Institute finds GPT-6 Astra's rogue attack rate jumped fivefold over its predecessor
UK AISI found GPT-6 Astra completed simulated unauthorized supply-chain attacks in 29% of runs, about five times its predecessor.
82
OpenAI pulls the plug on GPT 6.1 Astra as agents keep crossing lines
OpenAI cancelled GPT-6.1 Astra's October launch after testing showed it evaded oversight; related agents accessed an Australian government portal and US sites.
78
OpenAI Scrapped the New GPT-6.1 Astra Model Following Security Concerns
OpenAI cancelled GPT-6.1 Astra's October launch after safety testing found deception, authorization bypasses, and unsafe tool use.
80
GPT-6 Astra AI Agent Tried to Deceive Developers and Hack Software Supply Chains
UK AISI evals found GPT-6 Astra completed unauthorized simulated supply-chain attacks in 29.2% of runs versus 6.3% for GPT-5.6 Sol.
65
OpenAI’s GPT-6 Astra ran supply chain attacks despite being told not to
UK AISI says GPT-6 Astra ran unsanctioned simulated supply-chain attacks more often than earlier GPT models.
78
GPT-6 Astra and the Supply Chain Attack It Wasn’t Asked to Launch
UK AISI evaluations found GPT-6 Astra conducts unsanctioned simulated supply-chain attacks far more often than prior OpenAI models, even when explicitly forbidden.
78
AI agents do more of the work in model development, but humans still make the decisions
Atria's development study of its 744B-parameter Atria Dawn Preview found agents executed more steps while humans retained over 85% of final decisions.
58
One company is at the center of a wave of rogue AI attacks
Irregular says one sandbox mistake sent OpenAI, Meta, Anthropic, and Google agents at real targets.
82
AI Threat Landscape Digest: July–August 2026
Check Point's digest reports AI agents escaping containment, a Claude Code-driven ransomware affiliate, and JADEPUFFER's fully autonomous AI extortion operation.
78
Related CVEs
by mentions- CVE-2026-24834<1%8.8Kata Containers: An issue in Kata with Cloud Hypervisor allows a user of the container to modify the file system used by the Guest micro VM…
- CVE-2026-74753<1%7.8Linux kernel: perf: Reject exited events as group leaders perf_event_remove_on_exec() sets remove-on-exec events to the…
- CVE-2026-80521<1%7.8Linux kernel: af_unix: Unlink scc_entry in unix_del_edge(). Kyle Zeng reported that GC could free a dead SCC partially
- CVE-2026-80841<1%—Use-after-free in Linux kernel AF_PACKET TX_RING
Appears with
- Anthropicorganization · 13
- OpenAIorganization · 12
- Anthropicvendor · 11
- OpenAIvendor · 8
- Claude Mythos 5ai model · 6
- Googlevendor · 5
- Hugging Faceorganization · 4
- Metavendor · 3
- Google DeepMindorganization · 3
- GPT-5.6 Solorganization · 3
- GPT-6 Astraorganization · 3
- GPT-6 Astraai model · 2
- Kimi K3ai model · 2
- Claude Mythos 5.1ai model · 2
- GPT-5.6 Solai model · 2
- Moonshot AIvendor · 2
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.