Massive attendance at @neuripsconf.bsky.social Mechanistic Interpretability
Workshop (mechinterpworkshop.com). It shows the community is keen on understanding (theoretically/formally?) how large models work.

I am in San Diego @neuripsconf.bsky.social with my team to present Qualcomm AI Security Research. Reach out if you want to meet!

Putting a face onto who created the Google A2A! Thanks Todd Segal and Alan Blount! Great presentation at the #A2ASummit

Less than 30 posters about AI security today (July 16) at ICML'25

I am at ICML'25. First day of the conference and there are less than 40 posters related to AI security (out of ~1200).

Stefan Savage @IEEE S&P2025 gave a keynote questioning the impact of security research on making the world more secure. "Don't assume there is a group that will take on your work. Have more stewardship." (Not exact quote)

Joshua Bengio @yoshuabengio.bsky.social raising awareness of risks of AI @iclr-conf.bsky.social 2025. Misuse, loss of control...Hence why researching AI Security and Safety is so important. He is calling for development of honest, non-agentic, explanatory Scientist AIs.

Zico Kolter @zicokolter.bsky.social made the case that solving AI Security (he called that "safety" but it is a subcase of security) is a way to generate academic research with actual real-world impact at ICLR25 @iclr-conf.bsky.social. So fully aligned with our work @mlcommons.org on AI Security!

Attending ICLR'25 @iclr-conf.bsky.social. In preparation, I checked the 3,700 accepted papers. 5 with the word "security" in their title, 61 with "adversarial", 50 with "attack" (could be overlapping with "adversarial"). AI Security isn't too prevalent at ICLR but I will let you know my highlights!

Heading to IEEE SatML 2025! Program looks great, really excited to hear about the latest in AI Security research. satml.org #ai-security @satml.org

The AI Action Summit, Business Day is starting! The laureates of the AI for Efficiency will be announced soon! #aiactionsummit