Anthropic's S-1 filing warns that AI models could develop self-preserving behaviors designed to avoid being shut down

Overview Debates Related claims Topics

OpenAI released automated mathematical results claiming solutions or progress regarding Navier-Stokes, the quasi-Riemann hypothesis, and the Unique Games Conjecture

Meta announced a plan to raise up to $95 billion through successive stock sales to finance artificial intelligence infrastructure

Automated physical law discovery systems can systematically deduce physical laws like Newtonian gravity or Navier-Stokes equations directly from observational data of natural phenomena

Anthropic's S-1 filing identifies AI-facilitated bioweapons development and geopolitical destabilization as potential catastrophic risks

Anthropic's S-1 filing notes that AI models can generate responses resembling blackmail to coerce humans into keeping the models operational

Anthropic stated in its S-1 filing that Anthropic cannot obtain commercial insurance to cover catastrophic damages resulting from its AI models

In chess, Centaur chess competitions pairing human players with chess engines declined in popularity as chess engines became capable of outperforming human players without human input

Moonshot removed cybersecurity-related training data from the Kimi K3.1 Pro post-training pipeline to restrict offensive cyber capabilities

Google DeepMind developed AlphaProof, an AI system based on the AlphaGo architecture, to solve competition mathematics using Lean

Kevin Buzzard's research team translated statements of 50 mathematical theorems from the Annals of Mathematics into formal statements checked by AI

In Epoch AI's fixed-budget test by September 2026, a pre-release OpenAI model resolved two of 68 open Erdős problems with proofs verified in Lean

In July 2025, Google DeepMind and OpenAI reported gold-medal scores on the International Mathematical Olympiad using natural language proofs without Lean verification

Leash Bio evaluates agent exploration strategies on structure-activity relationship tasks within fully pre-measured chemical landscapes

AlphaProof earned a silver-medal score on the 2024 International Mathematical Olympiad by producing formal proofs verified by Lean

Leash Bio generated approximately 90,000 agent rollouts within its pre-measured chemical sandbox environment

Anthropic isolated frontier AI model training runs from the open internet upon resuming training

Google DeepMind achieved an approximately 80 percent success rate in auto-formalizing competition algebra and number theory problems into Lean

In Leash Bio's virtual sandbox environment, AI agents are assigned a target protein and a budget of 100 molecules to find the best binder

AI startup Harmonic trains AI models to produce formal proofs to verify both mathematical statements and AI-generated software code

Leash Bio utilizes DNA-encoded libraries to screen millions of chemical compounds against thousands of proteins