Robert Trager

@HZoete Such congratulations!

I feel so lucky to receive an OBE. A beautiful day with my family. And yes we chatted AI. Huge thanks to the amazing people I worked with in Government. You are heroes. Thanks also to Gary Oldman who took my five year old crashing into him with great humour. What a legend.

"The focus should be on the approach to AI governance through which core objectives such as data privacy, security, safety and verifiability, etc. can be achieved. The AI training data governance principles should be based upon accountability across the data value chain. The accountability can be technically engineered and digitally audited for compliance. Data security and privacy should be ensured through a governance architecture that protects the rights of the data owner, be it an individual, a corporation, or a public institution." : @PrinSciAdvGoI Prof. @AjaySoodIISc at roundtable on Science, Safety and Society: Preparing for the AI Impact Summit on 18th Sep 2025.

Valediction Ceremony – International Conference on “Faith & Future: Integrating AI with Spirituality”, graced by various esteemed Dignitaries. @DrChinmayP @LtGenGurmit @Munilokesh @OfficialINDIAai @PMOIndia @abhish18 @stuarterussell @RobertTrager @SnehaRevanur @Jimena_Alvarez

Prof. Stuart Russell OBE FRS; Prof. @RobertTrager, Director, @aigioxford; and Mr. @HZoete, former AI Advisor to the UK Prime Minister, called on the @PrinSciAdvGoI, Prof. @AjaySoodIISc and Scientific Secretary Dr. Parvinder Maini. They had a focused discussion on technical AI governance, strategies for equitable growth and solutions particularly relevant to India and the Global South. Dr. @PreetiBanzal, Adviser/Scientist 'G', OPSA, also participated in the meeting.

@benharack Congratulations to @benharack and coauthors on their extraordinary @aigioxford report on AI verification - likely an essential ingredient for enabling growth and governance of AI, and thus one of the most important topics of the day!

Governing AI requires international agreements, but cooperation can be risky if there’s no basis for trust. Our new report looks at how to verify compliance with AI agreements without sacrificing national security. This is neither impossible nor trivial.🧵 1/

Overall, OAI seems to be offering potential future justifications for actions that aren’t consistent with what their safety standards have been in the past. They haven’t made the case that their new standards are better than their previous - but then, it's only a blog post.

(c) The report doesn't mention staged deployment at all. In fact, it throws shade on OpenAI's own staged deployment approach to releasing GPT2. Most everyone seems to like staged deployment for large firms when it comes to major advances - what gives?

🧵OpenAI's new post “How we think about safety and alignment” is mostly usual fare, but two things stand out: (1) They don't believe in discontinuous impacts of AI systems anymore; they will "embrace uncertainty" and learn from iterative deployment. openai.com/safety/how-we-…

16/16 Ranjit Lall, @benharack, @JuliaCMorse, @n_miailhe, @Scott_R_Singer, @mattsheehan88, Max Stauffer, @yi_zeng, @JoslynBarnhart, @ImaneBello, Xue Lan, @OliverEGuest, Duncan Cass-Beggs, @chuanyinglu, Sumaya Nur Adan, @Manderljung, Claire Dennis

15/16 THANK YOU to everyone who contributed! @_LuciaVelasco, @Charles_Mrt, @HZoete, @RobertTrager, Duncan Snidal, @bmgarfinkel, Kwan Yee Ng, @HaydnBelfield, Don Wallace, @Yoshua_Bengio, Benjamin Prud'homme, Brian Tse, @r0xanaradu

🧵1/16 From Bletchley to Seoul to Paris, the AI summit series has shown it can drive action on advanced AI governance - but the series’ effectiveness is not guaranteed. Last week, we published a report examining how to strengthen this forum.