Michael L. Chen @miclchen

@RishiBommasani Thank you so much, Rishi!

Apply here by July 13: calcareers.ca.gov/CalHrPublic/Jo…

More roles coming out soon: stay tuned!

Come work with me in California state government! The California Governor's Office of Emergency Services (OES) is hiring a role for frontier AI policy implementation. Apply by July 13.

As Science Advisor, I'll also coordinate with AI governance leads across California's state government and facilitate cross-sector collaboration with the academic research community, the private sector, and community and nonprofit organizations.

At Cal OES, I look forward to working with frontier developers in reviewing assessments of risk from their internal use of frontier models. I'm also excited to help California prepare thoughtful, proactive response plans and playbooks for frontier AI risks.

I've joined California state government! As AI Science Advisor at the Office of Emergency Services (Cal OES), I'm advising senior leadership on frontier AI safety: critical safety incidents, AI and cyber defense, and risk from developers' internal deployment of AI.

@davidmanheim @MariusHobbhahn "What 2026 looks like" is much smarter than what I would have come up with if you asked me in 2021, I'll grant that

@jaeporeon @AndyMasley Instagram too

I wish more AI engineers working at Anthropic, Google, and OpenAI spoke to their gov affairs teams and knew what they actually said out here in DC. We are very far from solving the 'government affairs alignment problem'.

@AndrewCurran_ How much of this is distillation?

Reviewing AI 2027: > Mid 2025: Stumbling Agents > The world sees its first glimpse of AI agents. > Advertisements for computer-using agents emphasize the term “personal assistant”: you can prompt them with tasks like “order me a burrito on DoorDash” or “open my budget spreadsheet and sum this month’s expenses.” They will check in with you as needed: for example, to ask you to confirm purchases. Though more advanced than previous iterations like Operator, they struggle to get widespread usage. This is nonspecific enough that you could have written this about 2023. The next two paragraphs seem reasonable and very predictable from when AI 2027 was written (early 2025). > Late 2025: The World’s Most Expensive AI > OpenBrain’s latest public model—Agent-0—was trained with 10^27 FLOP. This seems like a considerable overestimate compared to Epoch estimates of the models that were actually released. > OpenBrain focuses on AIs that can speed up AI research. ... The same training environments that teach Agent-1 to autonomously code and web-browse also make it a good hacker. Moreover, it could offer substantial help to terrorists designing bioweapons, thanks to its PhD-level knowledge of every field and ability to browse the web. OpenBrain reassures the government that the model has been “aligned” so that it will refuse to comply with malicious requests. Sure, though this is also predictable or reflects already-existent research. The rest of this section seems accurate but also isn't much of prediction, so much as predicting the present of when AI 2027 was written. > Early 2026: Coding Automation > Overall, they are making algorithmic progress 50% faster than they would without AI assistants—and more importantly, faster than their competitors. Plausible but hard to say if we have good enough metrics to check this! This section seems broadly accurate but not hard to predict. Take what early-2025 AI is like, and make it a fair bit more capable. > Mid 2026: China Wakes Up > In China, the CCP is starting to feel the AGI. Maybe? I could see arguments for and against. > He sets in motion the nationalization of Chinese AI research, creating an immediate information-sharing mechanism for AI companies. It will escalate over the course of a year until all the best researchers merge into a DeepCent-led collective, where they share algorithmic insights, datasets, and compute resources with each other. Mid 2026 isn't over yet but I'd be surprised if this actually happens very soon. > So (CCP leadership wonder) should they act now and steal Agent-1? Or hold out for a more advanced model? If they wait, do they risk OpenBrain upgrading security beyond their ability to penetrate? Obviously very hard to validate what the CCP is thinking internally about model weight theft What's missing: the whole saga in early 2026 of not releasing an AI because of its powerful cyber capabilities, the US government reacting. Maybe that's sort of described in the February 2026 section. Reading on, some parts feel like they could be a self-fulfilling prophecy (e.g., The Geopolitics of Superintelligence).

@MariusHobbhahn Or to put it another way: many of its more dramatic predictions, it gets wrong; and many of the more dramatic and important things that actually happened in AI, it didn't predict.