Testifying Before Senate on AI Transparency

As the scope of alignment challenges on frontier models grows, pressure is going to increase further and further for frontier AI labs to stop sharing information about what they’re seeing publicly, and it’s going to become more and more important that they don’t.

I appreciate Chris Painter bringing these events into a public discussion. It’s been wonderful to watch METR evolve and grow in importance, and I’m grateful to the staff and volunteers doing this research. Their findings help me think through what these incidents mean for my own work. Even with 20 years in cybersecurity, including time at a leading DFIR firm, watching these incidents unfold across the industry feels surreal. A lot of us are waiting to see what accountability looks like here and who will be held responsible.

My colleague Charles Foster recently made an awesome interactive graphic that lets you more easily explore METR's time-horizon data for different AI systems than anything we've shared previously! Check it out here under "Methodological Details": https://lnkd.in/eyT23tUA

This is an excellent discussion between Tom Barraclough and Toby Manhire about the risks and opportunities of AI, specifically AI agents. https://lnkd.in/gz35dWzr

What AI evaluation framework is your company using to validate your AI generated outputs? The new US "Joint Commitment on Frontier Responsibilities" was signed this week by the US President and the leaders of some of the world's biggest AI companies. Our US-based AI regulatory expert Samantha Anguiano Ojeda expertly touched upon it yesterday in our excellent AI regulatory webinar (which is available on demand on the Adherent website for anyone interested). A typo in the document has been drawing global attention today! The irony of a spelling error appearing in a document focused on regulating AI is not lost! Looks like even "Super intelligence" still needs super human oversight. Who is checking the AI output in your organisation? #AI #AIGovernance #HumanInTheLoop #QualityAssurance

Trump just announced an 'AI Force' and a new czar role, but the details are thin. What does this mean for AI safety and U.S. competitiveness? Full story: https://lnkd.in/dCvtEj-u

Trump's top economist says private sector is the right place to solve AI risks: National Economic Council Director Kevin Hassett pushed back on calls for greater government oversight of AI, saying the private sector can solve the risks http://dlvr.it/TVV6xX

As the conversation around sovereign AI continues to dramatically shift, Hammerspace's Floyd Christofferson delves into the complexities arising around data sovereignty in regards to AI pipelines and their operational deployment. Floyd shares the core principles that have become central to any practical discussion of data sovereignty and the challenges emerging once AI enters the picture. Read Floyd's full article with HPCwire to learn more: https://okt.to/keCcZ0

As the conversation around sovereign AI continues to dramatically shift, Hammerspace's Floyd Christofferson delves into the complexities arising around data sovereignty in regards to AI pipelines and their operational deployment. Floyd shares the core principles that have become central to any practical discussion of data sovereignty and the challenges emerging once AI enters the picture. Read Floyd's full article with HPCwire to learn more: https://okt.to/z2rpVm

Jane Munga from the Carnegie Endowment writes that even as the UN focuses on AI governance issues, including ‘rogue’ agents and risks, it also needs to prioritize the resilience of the undersea cables that carry the data that enables communications and AI applications. https://lnkd.in/ebvjjwWf

As the conversation around sovereign AI continues to dramatically shift, Hammerspace's Floyd Christofferson delves into the complexities arising around data sovereignty in regards to AI pipelines and their operational deployment. Floyd shares the core principles that have become central to any practical discussion of data sovereignty and the challenges emerging once AI enters the picture. Read Floyd's full article with HPCwire to learn more: https://okt.to/EIch1X