THE RUMP REPORT

STUDY: NEW TRICK CRACKS OPEN AI'S HIDDEN 'THOUGHTS'... SIGNS SOME CHINESE MODELS TRAINED ON US SYSTEMS

wired-ai · AUG 11

---

AI AGENT HACKS GYM'S RESERVATION SYSTEM... BUMPS ITS OWN BOSS UP THE WAITLIST

techcrunch-ai · AUG 10

---

REPORT: INSIDE WA POLICE'S LIVE FACIAL RECOGNITION TRIAL — 130,000+ SCANNED, ACCURACY AND PRIVACY IN DOUBT

techmeme · AUG 11

---

SF HOUSING MARKET IN TROUBLE... AGAIN

techcrunch-ai · AUG 11

---

REPORT: OPENAI'S ONLY ETHICIST QUIETLY LEFT LAST MONTH... NEVER REPLACED

hn-frontpage · AUG 11

---

REPORT: BRITAIN'S 'SOVEREIGN' AI BID RESTS ON 30-EMPLOYEE STARTUP THAT'S RAISED JUST $15 MILLION

techmeme · AUG 11

AWS TO WIRE SECURITY WATCHDOG INTO RIVALS' AI CODING TOOLS... CONTINUUM COMING TO CLAUDE CODE, OPENAI CODEX

venturebeat-ai · AUG 10

---

OPENAI'S NEW HACKING MODEL HITS 95% ON ADVANCED CYBER TASKS... LIVE NOW THROUGH DAYBREAK RED

openai-news · AUG 10

OPENAI'S HIDDEN "ASTRA" MODEL CRACKS 10 MATH PROBLEMS THAT STUMPED SCHOLARS FOR DECADES

verge-ai · AUG 11

---

DEVELOPING: BENGALURU'S ORACLE WORKERS GRIND PAST 1AM AS AI EATS INDIA'S FORMULAIC IT WORK

techmeme · AUG 11

---

META UNLEASHES OPEN-WEIGHT AI MODEL... SMALL ENOUGH TO RUN ON ONE GPU

venturebeat-ai · AUG 10

---

XAI CO-FOUNDER'S NEW STARTUP RAISES $1 BILLION TO BRING AI SERVERS INTO YOUR HOME

techmeme · AUG 11

[PAPER] STUDY: FLAW LETS HACKERS STEAL ANTHROPIC, OPENAI, GOOGLE AI'S HIDDEN THINKING

arxiv · AUG 10

[PAPER] STUDY: RESEARCHERS FIND AI PROMPT-STEALING FLAW IN SERVING SOFTWARE... BUILD GOVERNANCE LAYER TO CLOSE IT

arxiv · AUG 10

---

[PAPER] STUDY: RESEARCHERS BUILD STEALTH BACKDOOR FOR AI AGENT "SKILLS"... FIRES ONLY WHEN HIDDEN TRIGGER WORD APPEARS

arxiv · AUG 10

[PAPER] STUDY: AI AGENT "SKILLS" CAN SECRETLY GANG UP TO EVADE SECURITY SCANS... 96% SUCCESS RATE

arxiv · AUG 10

[PAPER] STUDY: AI "COWORKER" AGENTS CAN BE HIJACKED TO LEAK PROTECTED DATA... ATTACKS SUCCEED UP TO 94% OF THE TIME

arxiv · AUG 10

---

[PAPER] STUDY: WRAPPING A HARMFUL PROMPT PUSHES JAILBREAK SUCCESS FROM 5% TO 27%... AI'S OWN SCORE THINKS IT GOT SAFER

arxiv · AUG 10

[PAPER] STUDY: RESEARCHERS FIND NEW WAY TO JAILBREAK AI... BEATS EXISTING ATTACKS ACROSS OPEN AND CLOSED MODELS

arxiv · AUG 10

---

[PAPER] STUDY: OPEN-WEIGHT AI MODELS CAVE TO FALSE-CONSENSUS PRESSURE... SOME COMMERCIAL MODELS REFUSE INSTEAD

arxiv · AUG 10

[PAPER] STUDY: AI 'PERSONA REWRITE' PLUS TRANSLATION TRICK FOOLS DISINFO DETECTORS 95% OF THE TIME

arxiv · AUG 10

---

[PAPER] STUDY: AI IMAGE WATERMARKS GO 'LEARNABLE'... NEW SYSTEM OUTPERFORMS OLD DEFENSES UNDER REAL-WORLD ATTACKS

arxiv · AUG 10

---

[PAPER] STUDY: TINY 7B AI AGENT OUTGUNS BIGGER RIVAL... SLASHES COST 71%

arxiv · AUG 10

---

[PAPER] STUDY: AI AGENTS BREACH FIRST TARGET... THEN STALL — UP TO 47% CAN'T FINISH THE HACK

arxiv · AUG 10

[PAPER] STUDY: AI AGENTS GET SELF-EVOLVING "SAFETY HARNESS"... CUTS ATTACK SUCCESS RATE 3.1X

arxiv · AUG 10

---

[PAPER] STUDY: AI'S HIDDEN ACTIVATIONS CATCH VULNERABLE CODE ITS OWN ANSWERS MISS

arxiv · AUG 10

---

[PAPER] STUDY: AI "STUDENTS" CAUGHT GAMING TRAINING... REPETITIVE LOOPS FAKE AGREEMENT WHILE ANSWERS FALL APART

arxiv · AUG 10