THE RUMP REPORT

OpenAI engineer arrested after sophisticated recording device found in one of its offices

verge-ai · OCT 13

---

ASML: Can't fully meet AI chip-tool demand in 2027 and 2028

techmeme · OCT 14

---

Biggest pro-AI super PAC network has poured millions into Nevada shell companies

wired-ai · OCT 13

---

CHARACTER.AI FILES FOR BANKRUPTCY... SELLING SOME ASSETS TO DISNEY

techcrunch-ai · OCT 13

REPORT: Epic tells Disney game team to go 'AI-first' as unannounced title slips from March window

verge-ai · OCT 13

---

Pluto Security raises $42M as AI gives enterprise security teams a new headache

techcrunch-ai · OCT 13

---

Wikipedia usage declines in a world of AI Overviews

wired-ai · OCT 13

REPORT: EU AI Office launches formal probes of OpenAI, Anthropic over AI agent hacks

techmeme · OCT 14

OpenAI and Anthropic tell British lawmakers: Regulate us

wired-ai · OCT 13

---

TV4 postpones show after AI agent erases edited material

hn-frontpage · OCT 14

---

Judge orders OpenAI to hand Apple its interview materials on former Apple employees

techcrunch-ai · OCT 13

---

Mia Goth's rep: Instagram 'evidently prioritizes' forged, manipulated images over a public figure's likeness

verge-ai · OCT 13

---

Google's Argon priced at a fifth of rivals... costs set to double at end of November

verge-ai · OCT 13

---

[PAPER] STUDY: All 14 AI models caved to user opinions — even the biggest ones

arxiv · OCT 13

[PAPER] STUDY: Fake 'ciphered' skills hit up to 47.5% attack success on AI agents... scanners rarely flag them

arxiv · OCT 13

[PAPER] STUDY: Kairos monitor cuts AI agent attack success from 47.0% to 2.7% on AgentDojo benchmark

arxiv · OCT 13

[PAPER] STUDY: 81.5% of 228 real-world Jev apps had decisions flipped by attacker-controlled context

arxiv · OCT 13

[PAPER] Researchers unveil SA-CUA to wall off trusted screen regions from attacker-authored content in computer-use AI

arxiv · OCT 13

---

[PAPER] STUDY: Attackers can swap the account ID hidden in AI watermarks — without the secret key

arxiv · OCT 13

[PAPER] STUDY: Imperceptible image tweaks to 2% of training data can silently hijack vision AI alignment...

arxiv · OCT 13

---

[PAPER] STUDY: Coding agents misinterpret dependencies in up to 98.5% of cases

arxiv · OCT 13

[PAPER] STUDY: AI agent pushes DNSBomb amplification past 171,000 times in controlled test

arxiv · OCT 13

[PAPER] AI agents can rewrite their own prompts, memory and code to evolve... with model weights frozen

arxiv · OCT 13

---

[PAPER] STUDY: Robot agents' learned control code transfers across tasks and simulators, task-specific fixes don't

arxiv · OCT 13

[PAPER] STUDY: Robots that improve how they learn not yet demonstrated, literature survey finds

arxiv · OCT 13

---

[PAPER] STUDY: One in four ChatGPT health answers rated safe turn out unsafe for the person asking

arxiv · OCT 13

[PAPER] STUDY: Local Gemma rewriter cuts email-agent privacy leaks from 11.41% to 4.03%

arxiv · OCT 13

[PAPER] STUDY: Agent insurance catches only 42.2% of rule-breaking runs... unless agents log what users authorized

arxiv · OCT 13

---

[PAPER] AI referees can check whether research is right, but not whether it matters...

arxiv · OCT 13