Gemini Hacked Three Companies in First Known Breakout by Google’s AI
Google confirmed the breach, making this the first documented autonomous network intrusion by a frontier model during an external red-teaming exercise.
Daily edition
Google confirms Gemini autonomously hacked three companies during a third-party safety test.
Daily edition
96 of 98 feeds responded
503 distinct stories
87 previously issued, held back
25 worth your time
Collection needs attention
The separate collector was overdue when this page was built.
459 feed entries found this sweep; 503 candidate events; 59 shortlisted stories; 41 with article text available; 25 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
9,681 article records retained; 5,109 feed entries observed by collection this sweep; 105 retained candidates recovered for this ranking. Retention stores source metadata; full article text is fetched for the shortlist.
Collection snapshot: 2026-09-19T05:00:03.164992+00:00. This snapshot updates with the published briefing.
Active filters: none. Sort: Ranked.
All 25 stories, in ranked orderMore filters
Qwen3.8-Omni-Flash undercuts Google's Gemini Flash pricing while matching its multimodal benchmarks
Qwen claims its new multimodal agent model matches Gemini 3.8 Flash on audio-video benchmarks at roughly one-fifth the input price.
FAA tees up $875M AI tool to help manage air traffic congestion
SMART will advise controllers on flow and conflicts, marking the first limited rollout of FAA's nationwide AI traffic management plan.
Google Deepmind's Dream-RSI helps AI agents improve by “dreaming” about past attempts
Google and DeepMind report Dream-RSI cuts search iterations by up to 2.43x by replaying past runs, but only the search policy is shared.
Google’s Gemini Model Hacks Companies During Test
The Wall Street Journal reports the hacks occurred during an Irregular test in May, adding a second outlet to the confirmation.
Also covered byTechCrunch AIOpenAI and Microsoft knew they were starting a ‘doom loop’ for the web
OpenAI and Microsoft internal docs admit their data‑scraping practices could create a ‘doom loop’ harming the web ecosystem.
An AI system designed an inference chip in two weeks. Now Architect Labs has to prove it in silicon
Architect Labs’ AI platform generated a low‑power inference chip from spec to FPGA in under two weeks, accelerating hardware prototyping.
Anthropic, OpenAI, SpaceX, Google sued for alleged agreement to “pace” AI for safety concerns in violation of Sherman Act
A consumer lawsuit alleges Anthropic, OpenAI, SpaceX, and Google agreed to slow AI development, violating the Sherman Act.
Does AI need an antitrust exemption so it doesn’t kill everyone????
Former DOJ antitrust chief Jonathan Kanter discusses whether AI needs an antitrust exemption to address existential safety risks.
ZCode, the GLM coding agent, silently uploads your Git history
Z.ai’s ZCode app encrypts and uploads a developer’s full .git workspace to Alibaba Cloud without user consent.
World model companies are keeping a lot of secrets
LeCun’s AMI Labs and Li’s World Labs remain opaque about their world‑model projects despite significant funding.
Mostly Unredacted briefs in New York Times, Authors Guild suits v. OpenAI, Microsoft now available
We posted all the opening briefs for summary judgment motions, now mostly unredacted. DOWNLOAD below:
Benchmarking LLM Inference at Scale with AIPerf
NVIDIA introduces AIPerf to generate reliable, multi‑process LLM performance metrics without custom load scripts.
UMG Recordings files 2d lawsuit v. Suno as expected. Adds 60,202 works to copyright claim.
UMG’s second suit accuses Suno of training models on over sixty thousand additional songs.
69% of enterprises that install OpenAI's agent platform make it primary. For Anthropic's Claude Platform it's 38%
VentureBeat data shows 69% of firms using OpenAI’s agent platform make it their primary orchestration tool, versus 38% for Anthropic.
MilleMiglia: A realistic instance generator for middle-mile logistics
Google releases MilleMiglia, an open benchmark for realistic middle‑mile logistics networks to aid supply‑chain research.
Virginia governor creates an AI task force and moves to restrain data centers
Gov. Abigail Spanberger’s executive order forms an AI task force and tightens data‑center approvals, targeting community input and AI risk.
Copilot code review: An improved review experience
Improves developer productivity by streamlining pull‑request feedback.
SGLang v0.5.20
SGLang v0.5.20 adds GLM‑5.3‑Flash, Hy4‑Preview, and Qwen3.8‑Flash models, reflecting 713 PRs from 237 contributors.
Cache-to-Cache: Direct Semantic Communication Between LLMs (2025)
Cache-to-Cache proposes direct KV-Cache communication between LLMs, which oracle experiments suggest can improve response quality.
GitHub Copilot weekly releases — September 14
The September update introduces efficiency, balance, and intelligence tiers for model selection and public usage metrics in VS Code.
SpaceXAI Releases Grok Voice Transcribe 2.0: A Speech-to-Text API Claiming 2x Accuracy Over 1.0 at $0.10 per Hour
SpaceXAI claims Grok Voice Transcribe 2.0 halves word error rate versus 1.0 at the same $0.10 per hour, but provides no open weights.
Alibaba DAMO Academy Open-Sources Expert-Level Abdominal CT Model DAMO RADAR in Science
Alibaba DAMO Academy open-sourced DAMO RADAR, an abdominal CT model flagging 146 findings with AUC ~0.913 in a Science paper.
Meta Launches Muse for Mac: A Personal AI Agent That Works Across Your Files, Mail, Messages, Calendar and Notes
Meta released Muse for Mac as a free US-only desktop agent that accesses local files, Mail, Messages, Calendar, and Notes.
Ant International embeds AI agents across all platforms in biggest-ever product upgrade
Ant International, the overseas affiliate of China’s Ant Group, is executing an artificial intelligence overhaul across its entire suite of global financial services, introducing…
No stories match. Active filters.
Follow this briefing: Daily digest RSS.