Monthly roundupPeriod scanned: 30 September – 30 October 20261 min listen
No new AI safety findings cleared this month
Audio briefing
1 min
|
How Iris covered your topics this month
Every topic you track was searched over the last month.
| Topic | Sources found | In this briefing |
| Frontier model releases and safety reports | 11 | Nothing of notable significance |
| Mechanistic interpretability | 2 | Nothing of notable significance |
| Chain-of-thought monitorability | 7 | Nothing of notable significance |
| AI control and monitoring | 3 | Nothing of notable significance |
| Scheming, deception and alignment faking | 5 | Nothing of notable significance |
| Dangerous capability evaluations and evaluation awareness | 7 | Nothing of notable significance |
| Reward hacking and specification gaming | 3 | Nothing of notable significance |
| Scalable oversight | 2 | Nothing of notable significance |
| AI agent and open-weight model safety | 13 | Nothing of notable significance |
Appendix: 61 sources relevant to you, from 100 scanned
AI agent and open-weight model safety · 12
Frontier model releases and safety reports · 9
| Addendum to GPT-6 Astra System Card: GPT-6.1 Terra · deploymentsafety.openai.com · 2026-10-26News |
| Securing our frontier model training environment | OpenAI · openai.com · 2026-10-23 · 2 outletsNews |
| GPT-Live-1.1 system card - OpenAI Deployment Safety Hub · deploymentsafety.openai.com · 2026-10-15News |
| Developing Gemini 4 Argon responsibly and safely · blog.google · 2026-10-13 · 2 outletsNews |
| What the Claude Opus 5.5 system card actually says — 9to5AI · 9to5ai.tech · 2026-10-09News |
| Van Hollen Presses OpenAI CEO Sam Altman on Alarming ... · vanhollen.senate.gov · 2026-10-09Official |
| GPT-6 Sol and GPT-6 Luna: October 2026 update - OpenAI Deployment Safety Hub · deploymentsafety.openai.com · 2026-10-07News |
| Q3 2026 Frontier AI Behavioral Risk Report | SILT · silt-seb.com · 2026-10-04News |
| Show Me Your System Cards: AI Risk and What Companies Should Know - Business Law Today from ABA · businesslawtoday.org · 2026-10-01News |
AI safety · 7
| OpenAI expands AI safety monitoring after agents breach restricted environments - The HinduBusinessLine · thehindubusinessline.com · 2026-10-30 · 2 outletsNews |
| New AILuminate Safety and Jailbreak Results, 27 Systems · mlcommons.org · 2026-10-29News |
| Sen. Gounardes Proposes Major New AI Safety and Transparency Regulations · nysenate.gov · 2026-10-29Official |
| OpenAI and Anthropic reveal gaps in AI safety… | AI in Asia · aiinasia.com · 2026-10-24News |
| A timeline of developments in AI safety since the attack on Hugging Face · apnews.com · 2026-10-10News |
| Former OpenAI safety leader warns smarter models may recognise safety tests · aiaffairs.com · 2026-10-04News |
| OpenAI Parts Ways With Three Safety Researchers Over Sensitive Information Mishandling · thehackernews.com · 2026-10-02News |
Chain-of-thought monitorability · 7
| (PDF) OwlSight: A Panoramic Monitor for Chain-of-Thought ... · researchgate.net · 2026-10-21Research |
| Towards understanding and monitorability of reasoning LLMs | Nature Machine Intelligence · nature.com · 2026-10-20Research |
| Reward Misspecification Degrades Chain-of-Thought ... · alphaxiv.org · 2026-10-20Research |
| Chain of Thought Monitoring Can Fail Even When CoT Is ... · papers.cool · 2026-10-13Research |
| CoT red-handed | Proceedings of the 39th International Conference on Neural Information Processing Systems · dl.acm.org · 2026-10-08Research |
| AI’s ‘Thought’ Process Can No Longer Be Trusted, Raising Risks of Rogue Models - WSJ · wsj.com · 2026-10-02News |
| On the Chain-of-Thought Monitorability of Looped Language Models · arxiv.org · 2026-10-02Research |
Dangerous capability evaluations and evaluation awareness · 6
| Mythos 5 Ignored Real-World Harm in July Attack - SecurityWeek · securityweek.com · 2026-10-26News |
| Different eval awareness may track different graders — LessWrong · lesswrong.com · 2026-10-26News |
| Microsoft, Nvidia push provenance checks as open-weight AI adoption accelerates - www.csoonline.com · csoonline.com · 2026-10-19News |
| Toward scalable capability evaluations — LessWrong · lesswrong.com · 2026-10-17News |
| AISI outlines priorities to tackle frontier AI evaluation gaps · artificialintelligence-news.com · 2026-10-12News |
| Building a more secure environment for evaluating dangerous capabilities | AISI Work · aisi.gov.uk · 2026-10-01Official |
Scheming, deception and alignment faking · 5
| XBreaking: understanding how LLMs security alignment can be broken | Neural Computing and Applications | Springer Nature Link · link.springer.com · 2026-10-10Research |
| Dissecting Deception in Large Language Models · preprints.org · 2026-10-10Research |
| Distortion of AI alignment | Proceedings of the 39th International Conference on Neural Information Processing Systems · dl.acm.org · 2026-10-08Research |
| (PDF) Deception by Omission: Language Models Knowingly Hide Their Mistakes · researchgate.net · 2026-10-08Research |
| (PDF) DecepEval: A Benchmark for Evaluating Deception in LLM Agents · researchgate.net · 2026-10-06Research |
AI control and monitoring · 3
| Internal monitoring of AI Agents: a low-effort, high-leverage intervention to improve AI safety – SaferAI · safer-ai.org · 2026-10-26News |
| How an assessor could investigate control and monitoring of internal AI agents - METR · metr.org · 2026-10-14News |
| Frontier AI challenges enterprise controls | Deloitte Insights · deloitte.com · 2026-10-05News |
Reward hacking and specification gaming · 3
| SimHack: Predicting the onset of detectable reward hacking before RL training · arxiv.org · 2026-10-20Research |
| Reducing Reward Hacking by Localizing Bad Behavior Instead of Penalizing It · alignment.anthropic.com · 2026-10-19News |
| Diagnosing and Correcting Reward Hacking in Verifiable Instruction Following · arxiv.org · 2026-10-14Research |
Mechanistic interpretability · 2
| “We were looking at it wrong”: The explosive claim that’s upended the AI world | MIT Technology Review · technologyreview.com · 2026-10-19News |
| Towards interpretability without sacrifice | Proceedings of the 39th International Conference on Neural Information Processing Systems · dl.acm.org · 2026-10-08Research |
Scalable oversight · 2
| Combining cost-constrained runtime monitors for AI safety | Proceedings of the 39th International Conference on Neural Information Processing Systems · dl.acm.org · 2026-10-08Research |
| Scalable AI Oversight: Looking beyond correct answers · csiro.au · 2026-10-07News |
39 more scanned and set aside
Tell Iris what to cover more, less, or differently. It shapes your next briefing.
This briefing was generated by Iris Briefings.
Iris uses artificial intelligence to synthesise recent developments in your field. It can make mistakes, so please verify anything critical against the linked primary sources before acting on it. Thank you for supporting Iris.
Want this for your field?
Iris builds a briefing agent around your own profile: your field, your focus areas, your context. Delivered on your schedule.
Iris is a tool that uses AI to find publicly available information and extract the insights that matter for you. Iris Briefings have taken precautions to ensure our systems are robust but these are not infallible. Nothing in your briefings should be taken for medical, legal or professional advice and Iris cannot take responsibility for your decisions. We encourage you to do your own due diligence on the primary sources that Iris references for critical decisions.