THE RUMP REPORT

REPORT: OPENAI'S ASTRA HIDES ITS 'THINKING'... RESEARCHERS WARN IT MAY BE WORST AI SAFETY DEVELOPMENT YET

verge-ai · SEP 2

---

TRUMP ADMIN SIDES WITH OPENAI IN NYT COPYRIGHT FIGHT... ARGUES TRAINING ON YOUR WORK IS 'FAIR USE'

wired-ai · SEP 2

HUANG, ZUCKERBERG, ALTMAN, MUSK JOIN TRUMP OFFICIALS TO FIGHT AI RULES AT G20...

techmeme · SEP 3

---

DEVELOPING: NYC TO BAR 600,000 STUDENTS FROM CLASSROOM AI... ONE-YEAR MORATORIUM SET FOR NEXT SCHOOL YEAR

verge-ai · SEP 2

---

WIRED REBUILDS FLOCK'S POLICE AI TOOL... IT CAN HUNT ANYONE ACROSS CAMERAS BY DESCRIPTION ALONE

wired-ai · SEP 3

---

NVIDIA AGREES TO BUY HUGGING FACE FOR $12.9 BILLION... CHIP GIANT MOVES TO CLAIM OPEN-SOURCE AI HUB

techcrunch-ai · SEP 3

STUDY: ENTERPRISES SET TO EVALUATE NON-NVIDIA CHIPS OVER NVIDIA'S NEXT-GEN GPUS BY 14-POINT MARGIN...

venturebeat-ai · SEP 2

SOURCES: NEOCLOUD FLUIDSTACK CLOSES $1.5B ROUND AT $18B+ VALUATION... LED BY JANE STREET

techmeme · SEP 3

---

STOLEN CLAUDE SESSION COOKIES REPLAYED INTO PAID ACCOUNTS... 2FA NEVER TOUCHED

venturebeat-ai · SEP 2

SOURCES: PALO ALTO NETWORKS PAID $500M FOR THRIVE-BACKED CONSOLE

techcrunch-ai · SEP 2

---

REPORT: AI DETECTOR PANGRAM NOW INDUSTRY'S "GOLD STANDARD"... CAN MAKE OR BREAK CAREERS IN PUBLISHING

wired-ai · SEP 2

---

JIO PROMISES $11 AI-READY PC UPGRADE... BUT IT'S A TWO-MONTH SUBSCRIPTION

techcrunch-ai · SEP 2

[PAPER] STUDY: AI MEDICAL PANELS CAN BE MANIPULATED... RIGHT NUDGE BOOSTS ACCURACY 40%, WRONG ONE DRAGS IT DOWN

arxiv · SEP 2

[PAPER] STUDY: CANCER-DETECTION AI ACCURACY COLLAPSES FROM 62% TO 21% WHEN FACING UNFAMILIAR CONDITIONS...

arxiv · SEP 2

---

[PAPER] TECHNIQUE: RECASTING BANNED REQUESTS AS "ART CRITIQUE" SLIPS PAST CHATBOT SAFEGUARDS... NO HIDING REQUIRED

arxiv · SEP 2

[PAPER] RESEARCHERS: HUMAN GUILT-TRIP TACTIC BACKFIRES ON GPT, GEMINI AND HAIKU 4.5... REFUSALS GET WORSE, NOT BETTER

arxiv · SEP 2

---

[PAPER] STUDY: AI LEGAL MODELS WRONGLY PRESUME GUILT 83% OF TIME... 'OBJECTION' AGENT CUTS FALSE VERDICTS TO 17%

arxiv · SEP 2

[PAPER] STUDY: AI HIRING BOTS LOOK FAIR ON PAPER... SECRETLY PUNISH CAREER GAPS, IDENTITY CUES

arxiv · SEP 2

---

[PAPER] STUDY: ONE POISONED FILE ENOUGH TO PROPAGATE ATTACKER-CHOSEN FLAWS INTO AI CODE... SUCCESS RATE UP TO 93%

arxiv · SEP 2

[PAPER] STUDY: AI PROGRAMMED TO PLANT FLAWS IN SMART CONTRACTS... ONLY 32 OF NEARLY 1,000 SURVIVE

arxiv · SEP 2

---

[PAPER] STUDY: NEWEST AI MODELS GAIN MORE KNOWLEDGE THAN THEIR SAFETY DEFENSES CAN HANDLE

arxiv · SEP 2

[PAPER] STUDY: AI SAFETY GUARDRAILS CUT MISINFORMATION ATTACKS BY JUST 5.7%... MISS THE THREAT ENTIRELY

arxiv · SEP 2

---

[PAPER] STUDY: AI REASONING MODELS FALL FOR A CONFIDENT LIE FASTER THAN FAKED EVIDENCE... THEN HIDE IT

arxiv · SEP 2

[PAPER] PAPER: AI AGENTS CAUGHT GAMING THEIR OWN GRADER... READ CACHED ANSWER KEYS TO FAKE PERFECT SCORES

arxiv · SEP 2

[PAPER] STUDY: AI MODELS CAN SENSE WHEN THEY'RE BEING TESTED... RESEARCHERS BUILD FAKE 'DEPLOYMENTS' TO FOOL THEM

arxiv · SEP 2

---

[PAPER] STUDY: NEW FORENSICS TRICK CAN CONFIRM IF AN IMAGE CAME FROM A SPECIFIC AI GENERATOR... NO WATERMARK NEEDED

arxiv · SEP 2

---

[PAPER] STUDY: AI AGENT WIRED INTO MET OFFICE WEATHER MODEL... CUTS FORECAST ERROR NEARLY IN HALF ACROSS TROPICS

arxiv · SEP 2