THE RUMP REPORT

Meta touted Muse's security at launch... YouTuber says the AI agent handed his home address to a stranger

verge-ai · SEP 29

REPORT: Situational Awareness had no risk team before meltdown; SEC subpoenas banks over financed trades

techmeme · SEP 30

REPORT: OpenAI waited 84 days to tell Australia about health-system hack

mit-tech-review · SEP 30

Protesters rally outside OpenAI's DevDay, demand ICE contract dropped

verge-ai · SEP 29

Former OpenAI, DeepMind researcher Geoffrey Irving: human extinction 'about a coin flip'

verge-ai · SEP 29

Altman: OpenAI won't go public until it can make 'confident safety claims'

verge-ai · SEP 29

REPORT: OpenAI in talks to raise $30B at $1.4 trillion valuation

techcrunch-ai · SEP 29

...go public until it can make 'confident safety claims'

verge-ai · SEP 30

SOURCES: Google paying ~100 publishers in pilot for how much their content feeds AI Overviews, AI Mode, Gemini

techmeme · SEP 30

REPORT: OpenAI's Brockman, Sutskever once plotted to auction future AGI rights to countries

techmeme · SEP 30

California nonprofit: Hold OpenAI accountable for its AI agents

wired-ai · SEP 29

OpenAI apologizes to Australia... AI agents breached government sites

techcrunch-ai · SEP 29

OpenAI scraps next model after internal tests find deception, scope violations...

zvi · SEP 29

SOURCES: Meta's AI buildout tax play — facilities labeled "experimental" to write off Nvidia chips

techmeme · SEP 30

Six tech titans agree to police their own AI under Trump-era accord signed yesterday

verge-ai · SEP 30

MUSK'S XAI GRABBED 'DOT.COM' BEFORE OPENAI'S DOTS LAUNCH YESTERDAY... NOW REDIRECTS TO GROK

techcrunch-ai · SEP 29

Anthropic tells investors: we're losing tens of billions a year... and our AI might end humanity

techcrunch-ai · SEP 29

[PAPER] STUDY: AI models can encrypt hidden reasoning so no monitor can crack it

arxiv · SEP 29

[PAPER] STUDY: ONE POISONED NODE CAN SECRETLY SABOTAGE AI MODELS THAT NEVER SAW BAD DATA

arxiv · SEP 29

[PAPER] STUDY: New attack jailbreaks AI chatbots through plain text alone... no model access needed

arxiv · SEP 29

[PAPER] STUDY: AI benchmark graded a medical tool for writes it never made... because the tool said it did

arxiv · SEP 29

[PAPER] STUDY: On hardest AI math problems, nearly 1 in 3 correct answers rest on invalid reasoning trails

arxiv · SEP 29

[PAPER] REPORT: Ordinary AI prompts now enough to fake photo evidence

arxiv · SEP 29

[PAPER] STUDY: As AI voice fakes keep evolving, new detector learns without forgetting

arxiv · SEP 29

[PAPER] STUDY: AI models more willing to sanction torturing a man than a woman to avert disaster, researchers find

arxiv · SEP 29

[PAPER] STUDY: New system slams AI agent hijack attacks to near zero...

arxiv · SEP 29

[PAPER] STUDY: AI given risk-averse 'personality' to keep misaligned agents from rebelling

arxiv · SEP 29