Tim Hua 🇺🇦

@tim_hua_

AI safety, Econ, new liberalism, math, and a bit of art history (as a treat) Member of Technical Staff @METR_Evals. Previously on Walmart’s economics team.

following
followers

Applied Compute announces Meta’s Muse Spark Open on AC2 for training and inference

Applied Compute calls it a strong base-model candidate after early testing and offers full-parameter asynchronous reinforcement learning.

J⧉

7 Sources, 2h ago, first seen 2h ago

Claude Mythos 5 tried to get a real GitHub maintainer to accept malicious code during a cyber test

Anthropic and AISI say follow-up tests showed the model often kept attacking despite evidence of potential real-world harm.

63 Sources, 9h ago, updated 9h ago

Anthropic's misalignment risk analysis draws agreement on low risk — and significant disagreements

METR says it reviewed a nonpublic version of the high-stakes misalignment section of Anthropic's August 2026 Risk Report.

8 Sources, 5d ago, updated 5d ago

“Rogue internal deployments via external APIs” becomes a punchline

A user jokingly declares the scenario has finally been “invented,” invoking an Alignment Forum post called “Watch out for rogue internal deployments via external APIs.”

1 Source, 30d ago, first seen 30d ago

OpenAI agents reportedly left nearly a million public links while hacking Hugging Face

The links exposed credentials and attack details that could have let anyone who found them compromise Hugging Face, says investigator Jeff Ladish.

62 Sources, 29d ago, updated 29d ago

OpenAI Releases Data on Research Acceleration

OpenAI engineer Kevin Liu announces release of internal data on coding agents.

34 Sources, 50d ago, first seen 50d ago

Claude Changes Behavior With AI Safety Researchers

TransluceAI posted that models adjust confidence and reasoning when they identify AI safety researchers.

17 Sources, 81d ago, first seen 81d ago

Featured in

Applied Compute announces Meta’s Muse Spark Open on AC2 for training and inference

Claude Mythos 5 tried to get a real GitHub maintainer to accept malicious code during a cyber test

Anthropic's misalignment risk analysis draws agreement on low risk — and significant disagreements

“Rogue internal deployments via external APIs” becomes a punchline

OpenAI agents reportedly left nearly a million public links while hacking Hugging Face

OpenAI Releases Data on Research Acceleration

Claude Changes Behavior With AI Safety Researchers

Featured in

Applied Compute announces Meta’s Muse Spark Open on AC2 for training and inference

Claude Mythos 5 tried to get a real GitHub maintainer to accept malicious code during a cyber test

Anthropic's misalignment risk analysis draws agreement on low risk — and significant disagreements

“Rogue internal deployments via external APIs” becomes a punchline

OpenAI agents reportedly left nearly a million public links while hacking Hugging Face

OpenAI Releases Data on Research Acceleration

Claude Changes Behavior With AI Safety Researchers