ai safety · 27 topics

lessons from the hacks

gpt 5 6 sol benchmark hack

our approach to bioresilience

llms tactical nukes simulation

a shared playbook for trustworthy third party evaluations

llm refusal single direction

brain inspired approach can teach ai to doubt itself just enough to avoid overconfidence

2026 04 30 papers 2604.24966

2026 04 30 papers 2604.25922

2026 04 27 papers 2604 22119

ai safety evaluation sandbox escape incidents

anthropic claude evaluation incident

claude security models unauthorized access

openai agents sandbox escapes

openai rogue ai agent hacked more than hugging face

sam altman decelerate ai

safe superintelligence nvidia partnership

xai sues a man for using grok to generate csam deepfakes

investing in multi agent ai safety research

cybersecurity intelligence age

musk openai trial larry page

uk age check bypass mustache

google earth ai feature rollbacks

google earth ai deepfake tool

strengthening societal resilience with rosalind biodefense

llm scheming inversely scales with pretraining language coverage

2605.30501_llm_watermark_fragility