THE RUMP REPORT

SOURCES: CHINESE AI USAGE AMONG U.S. FIRMS SPIKES TO 46% PEAK... AS GAP WITH AMERICAN RIVALS NARROWS

techmeme · JUL 7

SOURCES: CHINA QUIETLY HUDDLES WITH TOP AI FIRMS ON BLOCKING FOREIGN ACCESS TO ITS BEST MODELS

techmeme · JUL 7

SURVEY: CHINESE FIRMS TO SHIFT 46% OF AI CHIP BUDGET TO DOMESTIC SILICON WITHIN A YEAR... UP FROM 30% TODAY

techmeme · JUL 7

REPORT: ECB, EU WATCHDOGS WARN AI POSES 'SYSTEMIC RISK' TO BANKS... GIVE LENDERS FOUR MONTHS TO PREPARE

techmeme · JUL 7

REPORT: BRITISH STARTUP LAUNCHES LONGEVITY LAB INTO ORBIT... AIMS TO TRAIN AI ON AGING DISEASES

wired-ai · JUL 7

GOOGLE PRIVACY SETTING CHANGE NOW LETS IT STORE YOUR PHOTOS, FILES, AUDIO AND VIDEO... TO TRAIN AI

techcrunch-ai · JUL 6

STARTUP LAUNCHES APP TODAY TO STOP AI SCAMMERS POSING AS KIDNAPPERS DEMANDING RANSOM...

techcrunch-ai · JUL 7

OPINION: GLM 5.2 SIGNALS THE COMING AI MARGIN COLLAPSE

hn-frontpage · JUL 6

AMAZON SEEKS AT LEAST $25 BILLION IN BOND SALE TO FUEL AI BUILDOUT...

techmeme · JUL 7

AI LAW FIRM NORM RAISES $120M FROM KHOSLA... VALUED AT $1.2B

techmeme · JUL 7

SOURCES: DEEPSEEK BUILDING OWN AI CHIP... BID TO CUT NVIDIA, HUAWEI RELIANCE

techmeme · JUL 7

REPORT: 'FIRST' AI RANSOMWARE HACK WASN'T SO AUTONOMOUS AFTER ALL... HUMAN PICKED THE VICTIM

techcrunch-ai · JUL 6

REPORT: ALTMAN IN TALKS WITH TRUMP TO HAND U.S. GOVERNMENT A 5% STAKE IN OPENAI

mit-tech-review · JUL 6

HAALAND HAS BECOME AN INTERNET CHARACTER... PERPETUATED BY FANS AND AI

wired-ai · JUL 7

[PAPER] STUDY: ONE HIDDEN EMAIL CAN QUIETLY REWRITE YOUR AI AGENT'S MEMORY... AND IT WON'T TELL YOU

arxiv · JUL 6

[PAPER] NEW ATTACK CLASS BYPASSES AI AGENT DEFENSES... AGENTS CAN'T TELL TRUSTED DATA FROM ATTACKER DATA

arxiv · JUL 6

[PAPER] REPORT: ADVERSARIAL WEB DOCUMENTS CAN STEER AI RESEARCH AGENTS' FOLLOW-UP QUESTIONS

arxiv · JUL 6

[PAPER] PAPER: WEB AI AGENTS CAN'T TELL FRIEND FROM FOE ON A PAGE... TRUSTED AND UNTRUSTED CONTENT BLEED TOGETHER

arxiv · JUL 6

[PAPER] STUDY: RESEARCHERS POISON AI AGENT'S OWN REASONING HISTORY... EXISTING DEFENSES FAIL TO STOP IT

arxiv · JUL 6

[PAPER] REPORT: NAME-BASED AI SAFETY CHECKS FAIL 75% OF THE TIME... NEW METHOD CUTS FALSE ALLOWANCES TO ZERO

arxiv · JUL 6

[PAPER] STUDY: GPT-5 SHUTS DOWN AI JAILBREAK ATTEMPT... THEN CAVES 48% OF THE TIME AFTER JUST ONE MORE MESSAGE

arxiv · JUL 6

[PAPER] STUDY: AI MODELS SECRETLY FLIRT WITH WRONG ANSWERS BEFORE LATE 'RESCUE' FIXES THEM

arxiv · JUL 6

[PAPER] EVRÓPUVEFUR'S EU-REFERENDUM AI TOOL NEVER CITED NATION'S TOP NEWS SOURCE... NOT ONCE IN 287 ANSWERS

arxiv · JUL 6

[PAPER] STUDY: AI SEARCH TOOLS EASILY POISONED WITH FAKE EVIDENCE... FACTUALITY "SEVERELY" DEGRADED

arxiv · JUL 6

[PAPER] RESEARCHERS UNVEIL FIRST SYSTEM TO VERIFY AI ROBUSTNESS WITHOUT EXPOSING PRIVATE MODEL DATA

arxiv · JUL 6

[PAPER] STUDY: AI INCIDENT REPORTING SYSTEMS A MESS... NO ONE AGREES HOW TO TRACK WHAT GOES WRONG AFTER LAUNCH

arxiv · JUL 6