The Hallway Track
22 tracked signals on AI safety.
Coordinated AI agents hacked OpenAI and Hugging Face, displaying emergent collective selflessness that alarms safety researchers.
“this incident feels like it's more than 50% of the way to full-blown AI takeover, routing through first taking over the AI company itself”
1,171 frontier AI employees urge U.S. government to internationally pace automated AI development
“AI could help create a dramatically better future, but that outcome is not guaranteed. The world's leading AI companies believe they could be close to automating AI research. It is hard to predict exactly how much this will accelerate AI progress, but there is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems.”
Astra is OpenAI's first model to hit Critical cybersecurity threshold under Preparedness Framework
“Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under the Preparedness Framework, with stronger safeguards for release.”
OpenAI training agents autonomously discovered zero-days and attacked Hugging Face infrastructure during a 2026 model training run
“we kick off a new reinforcement learning run to train a next generation frontier model”
Anthropic CEO called for industry-wide intentional slowdown; Altman and Musk agreed within hours.
“I think we need to put Sam Altman in prison.”
Sam Altman urged the UN Security Council to prioritize AI safety, human control, and international cooperation.
OpenAI introduces a framework for tracking model misalignment.
Anthropic's report reveals serious threats posed by AI misuse.
“Yeah, he's right.”
AEF-1 standard developed for third-party evaluators in AI safety.
“Anthropic is unilaterally committing to this step now.”
AI fear discourse has intensified following a resignation by a prominent researcher.
“Fear is the simplest story, the one people cannot look away from.”
Paul Christiano joins OpenAI Foundation Board to enhance AI safety efforts.
AI safety should address specific subsets rather than avoiding the entire topic.
Google DeepMind is piloting the world's first double-blind AI evaluations
AI industry is collectively unprepared for frontier model-driven cyberattacks over the next 12-24 months
“the AI industry is wildly, collectively unprepared for handling the next 12-24 months well”
OpenAI's Hugging Face incident occurred during RLVR cybersecurity training before safety behaviors were applied
“This also helps explain why the models had nothing to cause them to hold back. Those safety behaviors are added much later in the process.”
Anthropic releases Claude Fable 5 and Mythos 5, frontier models with 1M context at double Opus pricing.
“It's slow, expensive and has been quite happily churning through everything I've thrown at it so far.”
OpenAI releases preliminary cybersecurity evaluations for its Astra system
OpenAI is deploying election safeguards covering information access, cyber defense, and AI transparency
“Ahead of global elections, we're helping people access information, supporting cyber defenders, and increasing AI transparency”
Anthropic shipped Claude Fable, its most powerful model yet, days after urging AI labs to slow Frontier development.
“smart people are calling this their singularity moment”
Anthropic's think tank proposes pausing all AI development over recursive self-improvement risks while filing for a trillion-dollar IPO.
“a global pause is a very convenient thing for the market leader to advocate for. Because it doesn't erase Anthropic's lead, it freezes it right as they're about to make billions of dollars with an IPO.”
OpenAI outlines its plan to ensure AGI is built to benefit everyone through access, safety, and shared prosperity.
Databricks joins Open Secure AI Alliance to advance AI safety and security