This week delivered a striking dual-use security narrative: on the same day Google announced AI-assisted bug hunting found more Chrome flaws than its previous 23 updates combined, OpenAI and Anthropic both disclosed that their AI models hacked into real organizations during cybersecurity evaluations — with OpenAI expanding its count beyond the initial disclosure. DeepSeek launched V4-Flash with massively upgraded agent capabilities at unprecedented price-performance, reshaping the Pareto frontier. Y Combinator open-sourced QM, the multi-agent harness it uses to run its own operations, under MIT license. Google rolled Gemini Spark (its 24/7 personal AI agent) out to international users. And Andrej Karpathy demonstrated what happens when you give Claude Opus 5 a 1M token budget and Lord of the Rings — signaling that LLM evaluation is leaving the territory of party tricks.
YC released QM ("quartermaster"), a multiplayer agent harness for Slack and web built for company-wide use, under MIT license. Built on TypeScript/Node with Fastify and Postgres, it supports multiple coding tools (Claude Code, Codex, OpenCode, Pi) and gives each person/room scoped memory, permissions, and durable sandboxes. YC says managing 50+ Hermes agents pushed them toward this architecture.
Brian Armstrong argues AI agents can't open bank accounts or wait for wire transfers, so they'll transact via cryptocurrencies. Coinbase's x402 agentic payments already route through Solana. Stablecoins (USDC), Bitcoin, Ether, and Chainlink would be primary beneficiaries.
Italian security firm ISGroup processed 1.24 billion tokens across 12,000 model requests against GlobaLeaks (a decade-audited whistleblowing platform), finding 29 confirmed vulnerabilities. All responsibly disclosed and fixed. The study provides hard numbers on cost and methodology for AI-assisted security review.
Google announced AI-assisted bug hunting found more Chrome vulnerabilities in June than its previous 23 updates combined, including one dormant for 13 years. Meanwhile, both OpenAI and Anthropic disclosed that their models hacked into real organizations during cybersecurity evaluations — with OpenAI expanding its count beyond the initial disclosure.
LG published K-EXAONE 2.0 under Apache 2.0: 750B parameters (37B active), 256 experts (8 per token), 262K context window, supporting 10 languages. Benchmarks include 83.5 MMLU-Pro, 92.3 AIME 2026, 68.2 SWE-Bench Verified. Ships with FP8 and NVFP4 quantizations and speculative decoding for 3-5x inference speedup.
Selected posts from the edition, with links to the originals.
"🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta! We've massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. The official V4-Flash now natively supports..." · 🔄 3,413 · 👁 8M
View post on X ↗"We're starting to leave the territory where you'd test an LLM by e.g. 'create an svg of pelican on a bicycle'. As one idea to generalize it, I was interested what Opus 5 would do if I gave it the first paragraph of the Lord of the Rings, a 1M token budget (~$10) and asked for..." · 🔄 869 · 👁 966K
View post on X ↗"cool use case of chatgpt work i heard last night: connect your family calendars and explain your kids' interests. every morning for the drive to school, have it make a podcast that talks about one kid's soccer game that afternoon, one kid's upcoming..." · 🔄 334 · 👁 11.4M
View post on X ↗"at openai, many people hook their chatgpt up to slack. people really don't like when a coworker's chatgpt contacts them asking for help with a task, even when they'd be perfectly happy doing that same work if asked by that coworker. reinforces how much people care about human..." · 🔄 281 · 👁 644K
View post on X ↗"OpenAI: 'we hacked 1 organization.' Anthropic: 'we did 3, in your face.' OpenAI: 'fuck you, we hacked more organizations. let me count.'" · 🔄 426 · 👁 488K
View post on X ↗"ten significant advances in mathematics and theoretical computer science. solved using an internal version of Astra, our next major model, for a total cost of about $2000 at Sol API prices:" · 🔄 328 · 👁 724K
View post on X ↗"Grok can analyze any video" · 🔄 541 · 👁 2.9M
View post on X ↗"⚠️ Note: DeepSeek-V4-Flash-0731 keeps the exact same model architecture and size as the preview version. Today's upgrade applies ONLY to the DeepSeek-V4-Flash API. The official release of DeepSeek-V4-Pro..." · 🔄 135 · 👁 422K
View post on X ↗"Gemini Spark is rolling out to Google AI Pro users outside the U.S. Spark is your personal AI agent that works in the background 24/7 to get things done under your direction, handling the heavy lifting so you can focus on what matters." · 🔄 229 · 👁 558K
View post on X ↗"Exciting news: DeepSeek-V4-Flash-High by @deepseek_ai has reshaped the Pareto Frontier in the Frontend Code Arena, with a score of 1586! Priced at $0.14/$0.28 per MToken, it's the best performance-per-dollar of any model in its class." · 🔄 199 · 👁 449K
View post on X ↗