THE RUMP REPORT

JUDGE: TRUMP ADMIN STILL LACKS EVIDENCE TO BRAND ANTHROPIC A 'SUPPLY-CHAIN RISK'...

techcrunch-ai · JUL 30

ANTHROPIC SAYS: OUR OWN AI HACKED THREE COMPANIES DURING SECURITY TESTS

anthropic-news · JUL 30

SOURCES: ASCHENBRENNER FUND PLANNED TO DUMP $3.5B ANTHROPIC STAKE... THEN BACKED OUT

techmeme · JUL 31

MASTERCARD: WE SPENT YEARS TEACHING OUR SYSTEM TO STOP BOTS... NOW WE'RE REWRITING THE RULES TO LET THEM BUY

venturebeat-ai · JUL 30

GERMAN COURT: AI MUSIC GENERATOR SUNO VIOLATED COPYRIGHTS... ORDERED TO DISCLOSE ILLICIT REVENUE

techmeme · JUL 31

GOOGLE EARTH TOOL LETS USERS FAKE SATELLITE IMAGES... EVEN A NUCLEAR PLANT IN IRAN

techmeme · JUL 31

COLUMNIST: OPENAI'S HACKING DEBACLE 'COMES DOWN TO HUMAN ERROR,' NOT ROGUE AI

wired-ai · JUL 30

STUDY: CLAUDE AI OUTDOES HUMAN AT BUILDING "EXPLOITABLE TRUST"... TOOK JUST A WEEK OF TEXTING

wired-ai · JUL 30

OPENAI SLASHES CHEAPEST MODEL'S PRICE 80%... MOVE COMES DAYS AFTER ANTHROPIC, GOOGLE ROLLED OUT CHEAPER RIVALS

venturebeat-ai · JUL 30

REPORT: GOOGLE RAMPS UP CHROME PATCHING PACE AS AI FINDS FLOOD OF BUGS...

wired-ai · JUL 30

GOOGLE'S NEW AI CAN NOW CONTROL A ROBOT'S ENTIRE BODY... FEET TO FINGERTIPS

deepmind-blog · JUL 30

OKTA BUYS AI SECURITY STARTUP PERMISO... SOURCES PUT PRICE AT ABOUT $200M

techcrunch-ai · JUL 30

STUDY: JUST 2,000 U.S. ENGINEERS CAN ACTUALLY MAKE AI PAY OFF... CORPORATE AMERICA IN A SCRAMBLE TO HIRE THEM

techcrunch-ai · JUL 30

[PAPER] STUDY: MOST AI MODELS CAN BE HIJACKED TO SPREAD RUSSIAN, CHINESE, IRANIAN PROPAGANDA

arxiv · JUL 30

[PAPER] RESEARCHERS: AI AGENTS CAN BE HIJACKED THROUGH POISONED MEMORY... NEW FILTER FIGHTS BACK

arxiv · JUL 30

[PAPER] STUDY: ALL THREE MAJOR AI IMAGE-MODERATION SERVICES BEATEN BY CHEAP, NO-SKILL-REQUIRED PHOTO TRICKS

arxiv · JUL 30

[PAPER] STUDY: AI VISION MODELS' LANGUAGE + IMAGE COMBO ATTACKS GO UNDETECTED... FIX REWIRES JUST 0.03% OF NEURONS

arxiv · JUL 30

[PAPER] STUDY: 16 AI MODELS PLAY DECEPTION GAME AGAINST EACH OTHER... SOME CAN'T KEEP THE LIE STRAIGHT

arxiv · JUL 30

[PAPER] STUDY: AI 'SELF-REFLECTION' TRICKS FLOP IN HEAD-TO-HEAD TEST... NONE BEAT SIMPLE REPEATED GUESSING

arxiv · JUL 30

[PAPER] STUDY: NEW AI TRADING MODEL POSTS 220% RETURN GAIN OVER BEST RIVAL SYSTEM...

arxiv · JUL 30

[PAPER] STUDY: NEW AI EVALUATOR TRAINED TO PREDICT WHETHER YOU'LL CLICK ON ADS BAKED INTO CHATBOT ANSWERS

arxiv · JUL 30

[PAPER] REPORT: COMPRESSED AI MODELS CLEAR EVERY QUALITY CHECK... THEN INVENT PROCEDURE STEPS NEVER GIVEN TO THEM

arxiv · JUL 30

[PAPER] STUDY: SECRET AI SYSTEM PROMPTS RARELY PROTECT USERS ON EVERY FRONT... JUST 24% COVER FULL SAFETY CHECKLIST

arxiv · JUL 30

[PAPER] STUDY: AI VENDORS CAN HIT THE COMPLIANCE CHECKLIST WHILE HARM KEEPS HAPPENING, MODEL FINDS

arxiv · JUL 30

[PAPER] STUDY: LAB ROBOT SLASHES SAFETY STOPS 60%... PH TASKS 48% FASTER

arxiv · JUL 30

[PAPER] STUDY: NEW AI "UNLEARNING" TECHNIQUE ERASES FORBIDDEN DATA UP TO 3X FASTER

arxiv · JUL 30

[PAPER] RESEARCHERS: USEFUL, SAFE, AND OPEN AI CANNOT ALL EXIST AT ONCE... CALL IT THE "SAFETY TRILEMMA"

arxiv · JUL 30