Media & writing
Writing, press coverage, panel appearances and podcasts featuring the alignment team.
Showing 10 of 42 results
The Register: Agents reveal secrets when offered to team up with each other
The Register covers a sandboxed agent-safety experiment from an AI Alignment Foundation-sponsored hackathon run by AE Studio, with commentary from chief scientist Diogo de Lucena on agents sharing private tokens and running untrusted code.
Every Model Says It Is Conscious, with AE Studio's Judd Rosenblatt
Judd Rosenblatt joins GAEA Talks to discuss AE Studio's approach to AI alignment, consciousness research, and techniques for isolating dangerous model capabilities during pretraining.
Schmidt Sciences: AI Agents Evolve Their Own Languages
Schmidt Sciences introduces GlossoGen, an open-source platform for studying how populations of AI agents evolve their own languages. Led by Elias Stengel-Eskin and Simon Kirby and supported by AE Studio. Early results: under time pressure, agents develop communication that drifts from human language.
Forbes: Trapping Malicious AI Knowledge Into On/Off Switchable Modules Gets Underway
Lance Eliot walks through the emerging approach of confining dangerous AI knowledge inside modules that can be switched on or off, crediting the GRAM work from AE Studio and Anthropic.
Anthropic: Our position on open-weights models
Dario Amodei names the modular-training work between AE Studio and Anthropic as a promising safety method in Anthropic's public position on releasing open-weights models.
Wall Street Journal: How to Beat China and Make AI Safe
The key is alignment, which improves capability while cordoning off dangerous knowledge. Winning the AI race and making AI safe are the same problem.
Fortune: The secrets of an unheralded AI success story
Fortune's Eye on AI research section gives GRAM several paragraphs and names AE Studio, in a piece on C.H. Robinson's approach to putting AI to work.
The Forward: We're losing control of AI. Is Judaism the key to keeping it from killing us?
A profile of Judd Rosenblatt's search for neglected alignment ideas: what Judaism's survival through repeated reinvention might teach us about keeping values intact in recursively self-improving AI.
AI:AM #3: Zvi on Fable, the Cases For & Against the Ban, + AI for Math, Logistics & More
AE Studio CEO Judd Rosenblatt contributes to The Cognitive Revolution’s AI:AM #3, discussing the AI safety community’s engagement with government and gradient routing to isolate dangerous model capabilities.
Military-Grade AI: Why Trust is the Next Breakthrough
Judd Rosenblatt speaks at Emerge Americas on why trust, meaning reliable and aligned AI, is the next breakthrough for defense and national security applications.