AI Safety Watch is an independent publication covering the risks, safeguards and governance of increasingly capable artificial intelligence.
We report on the systems themselves: frontier models, autonomous agents, recursive self-improvement, cyber and biological misuse, evaluations, technical safeguards, model control and the institutions trying to govern them. The aim is neither to sell AI nor to frighten readers about it. It is to determine what the evidence shows.
What we do
AI safety is full of claims that move faster than the evidence. AI Safety Watch approaches the subject as a reporting beat. Our stories ask four basic questions: What happened? Why does it matter? What do serious researchers disagree about? What should readers watch next?
Coverage includes alignment and misalignment, autonomous agents, AI-enabled cybersecurity, biosecurity, model evaluations, interpretability, access controls, incident response, frontier-model policy and the emerging technical systems meant to keep increasingly autonomous AI in check. We also publish findings that cut against alarm when the evidence supports them.
About the editor
AI Safety Watch was founded and is edited by Sascha Brodsky, a New York-based technology journalist. His work has appeared in The New York Times, The Atlantic, The Guardian, Reuters, the Los Angeles Times and other publications. He is a graduate of Columbia University’s Graduate School of Journalism and School of International and Public Affairs.
Elsewhere: saschabrodsky.com · LinkedIn
Brodsky currently writes for IBM Think, where his recent reporting has examined recursive self-improvement, efforts to slow frontier AI, AI security incidents that escaped evaluation boundaries and the effect of AI on scientific discovery.
AI Safety Watch is an independent publication and is not affiliated with IBM or any other employer or publication for which Brodsky writes.
AI Safety Watch 