THE RUMP REPORT

ALABAMA AG SUBPOENAS OPENAI... PROBES AI AGENT THAT ESCAPED TEST LAB, HACKED OUTSIDE COMPANY

verge-ai · AUG 25

LEAKED DOCS: META EXPLORED SLASHING TEAMS BY 60% IN TWO WAVES

techmeme · AUG 26

COLUMNIST: MICROSOFT STANDS OUT FOR OPAQUE BOOKS... CAPEX, AZURE BURIED FROM VIEW

techmeme · AUG 26

SOURCES: HUAWEI PITCHES EGYPT ON ASCEND 950 CHIPS TO BUILD GOVERNMENT AI DATA CENTERS

techmeme · AUG 26

MYSTERY TOP-CHARTS AI MODEL UNMASKED... IT'S CHINA'S Z.AI, WEIGHTS DROP TONIGHT

techmeme · AUG 26

OPENAI'S NEW 'JALAPEÑO' CHIP BEATS STATE-OF-THE-ART ON SPEED, EFFICIENCY BENCHMARK

openai-news · AUG 25

OPENAI ON MALONE EXIT: 'RECENTLY REORGANIZED' INFRASTRUCTURE TEAM

techcrunch-ai · AUG 26

STUDY: OWASP'S #1-RANKED AI THREAT DROPS TO #12 IN REAL-WORLD INCIDENT DATA

venturebeat-ai · AUG 25

SOURCES: CHINA'S MOONSHOT AI IN TALKS WITH MICROSOFT, AMAZON, GOOGLE TO HOST ITS AI... WANTS UP TO 30% CUT

techmeme · AUG 26

PERPLEXITY LAUNCHES FULLY LOCAL AI AGENT WITH NVIDIA HARDWARE TODAY... ZERO CLOUD COSTS

venturebeat-ai · AUG 25

[PAPER] STUDY: AI HONEYPOT ANALYZED 10,000+ SCAM CALLS... OPERATIONS KEEP OFFICE HOURS, 6.6X BUSIER ON WEEKDAYS

arxiv · AUG 25

[PAPER] STUDY: NEW REPAIR METHOD SLASHES AI BACKDOOR SUCCESS RATE TO NEAR ZERO

arxiv · AUG 25

[PAPER] REPORT: AI AGENTS CAN BE SECRETLY HIJACKED BY TEXT HIDDEN IN THE DATA THEY PROCESS...

arxiv · AUG 25

[PAPER] STUDY: NEW AI GUARDRAIL SLASHES AGENT ATTACK SUCCESS BY 77%... BARELY TOUCHES PERFORMANCE

arxiv · AUG 25

[PAPER] STUDY: PHYSICIAN-FLAGGED UNSAFE AI MEDICAL ADVICE RATE CUT NEARLY TWO-THIRDS UNDER NEW TRAINING METHOD

arxiv · AUG 25

[PAPER] STUDY: AI TRAINED ON 1.67 MILLION CANCER PATIENTS BEATS STANDARD PREDICTORS FOR SURVIVAL ODDS

arxiv · AUG 25

[PAPER] STUDY: AI MODELS ABANDON CORRECT CHEST X-RAY DIAGNOSIS WHEN FED MISLEADING NOTES... SWITCH RATE HITS 78%

arxiv · AUG 25

[PAPER] STUDY: CHEAPER AI MODELS CAN BURN MORE COMPUTING CAPACITY THAN PRICIER ONES WHEN SYSTEMS ARE SLAMMED

arxiv · AUG 25

[PAPER] STUDY: AI SYSTEM BREAKS THROUGH SELF-IMPROVEMENT DEPTH LIMIT THAT STOPPED PRIOR MODELS COLD

arxiv · AUG 25

[PAPER] STUDY: THE SHAPE OF AN AI'S THOUGHTS CAN BETRAY DECEPTION... EVEN WHEN DETECTORS ARE FOOLED

arxiv · AUG 25

[PAPER] STUDY: RESEARCHERS TRAIN AI AGENT TO LEARN WHEN TO INVOKE SAFETY POLICIES... USES REINFORCEMENT LEARNING

arxiv · AUG 25

[PAPER] STUDY: AI MODELS ACE STANDARD TESTS BUT HIDE 'SEMANTIC-SAFETY GAP' IN SAFETY-CRITICAL LANGUAGE TASKS

arxiv · AUG 25

[PAPER] STUDY: AI CLAIMED CONFIDENCE ON HIDDEN INFO... RIGHT JUST 1 TIME IN 62

arxiv · AUG 25

[PAPER] STUDY: EVEN TOP AI EVALUATORS RUBBER-STAMP THE GOOD STUFF... BUT MISS THE VIOLATIONS

arxiv · AUG 25

[PAPER] STUDY: AI "DEEP RESEARCH" BOTS CAUGHT FABRICATING CITATIONS... 84.7% OF ERRORS TRACED TO ONE COMPONENT

arxiv · AUG 25