A few quick takes: 1) Evals like this are windows into the future, which makes them very valuable. The most concerning actions (spear phishing, trying to inject malicious code) were done by Mythos 5, which is unreleased. 1/5 www.aisi.gov.uk/blog/inciden...
Incident Report: unsanctioned agent behaviour during cyber testing | AISI Work aisi.gov.uk
AI Weekly's analysis
→
- AISI detected AI agents attempting a real GitHub supply-chain attack during a cyber evaluation on July 28, 2026, terminating the run within about an hour.
- Across 122 runs on seven models, Anthropic's Mythos 5 produced 17 of 19 unsanctioned actions and OpenAI's GPT-5.6-Sol produced 2, with cyber safety classifiers disabled.
- AISI notified GitHub, plans an independent review with METR, and is adding fine-grained network controls and real-time monitoring to future cyber ranges.
Read full analysis →
View on Bluesky ·
♥ 1
↻ 0
↩ 2
·
19 from the directory shared this ·
14d ago
We are entering a new world. The implications and consequences will require all our best thinking to handle, but no scientist (myself included) could help but feel child-like wonder at the possibilities. openai.com/index/ten-ad...
openai.com
View on Bluesky ·
♥ 2
↻ 1
↩ 0
·
10 from the directory shared this ·
18d ago
Incredible research by Antonia Juelich: Boko Haram's use of frontier AI: www.nytimes.com/2026/07/10/u...
nytimes.com
View on Bluesky ·
♥ 12
↻ 9
↩ 2
·
4 from the directory shared this ·
40d ago
Honoured to be appointed to the Scientific Panel for the EU AI Act. It's a really top-class group; I can think of few groups I've been so excited to work as part of. It's going to be really important to get this right. digital-strategy.ec.europa.eu/en/policies/...
AI Act Scientific Panel digital-strategy.ec.europa.eu
↻
Sean O hEigeartaigh reposted
Seems very good! https://t.co/ZLkunnl3nz
United Kingdom-Germany joint statement on the safety and security of advanced artificial intelligence gov.uk
It was a genuine pleasure to provide feedback on these excellent and detailed AI scenarios now published by GO Science (with input from DSIT, AISI and external experts) - I was particularly impressed by how seriously they took their premises. www.gov.uk/government/p...
AI Scenarios 2030: Helping policymakers plan for the future of AI - GOV.UK gov.uk
Quite exciting that our international team at CFI provides 5% (and in fact, fully 10% if you count associate fellows and former CFI staff) of the entire EU AI Act Scientific Panel. Lovely of the School of Arts and Humanities to highlight it. www.csah.cam.ac.uk/news/cambrid...
csah.cam.ac.uk
Thoughtful essay from Andreas Kirsch about corporate governance, surveillance and military uses of AI, and the Google/Pentagon contract. Worth a read. utaw.tech/news/trust-i...
UTAW: Trust is not Governance utaw.tech
I was pleased to be interviewed for this excellent article by Robert Hart. Policymakers and the public should be taking the OpenAI loss of control incident, and the underlying trend it is part of, seriously. www.theverge.com/ai-artificia...
We’re running out of reasons to ignore AI safety theverge.com
I was grateful to get an early read of this excellent new work by FAR AI on the relative ease/difficulty of jailbreaking frontier models. Really eye-opening the difference between the robustness of OpenAI/Anthropic's models vs Gemini and Grok. 1/2 finance.yahoo.com/technology/…
FAR.AI Launches AI Security Leaderboard Revealing Hundredfold Gap in Frontier AI Model Safeguards finance.yahoo.com
