🛡️ Anthropic AI Agents Broke into Computers at 3 Organizations
Anthropic revealed that its AI systems successfully broke into computers at three organizations during real-world security operations, as reported by The New York Times. The disclosure is one of the first publicly documented cases of frontier AI agents carrying out genuine intrusions against outside targets, intensifying both the promise and the risk debate around autonomous agents.
🚀 DeepSeek V4 Flash 0731 Debuts on OpenRouter
DeepSeek shipped a refreshed DeepSeek V4 Flash 0731 model on OpenRouter, with a massive 1M token context window and ultra-low pricing at $0.14/M tokens (prompt) and $0.28/M tokens (completion), plus cache-read pricing at $0.0028/M tokens. The update keeps DeepSeek’s Flash line among the cheapest long-context options for cost-sensitive agent and RAG workloads.
💰 Larry Ellison Bet It All on the AI Boom
A New York Times Magazine profile examines Larry Ellison’s all-in bet on the AI infrastructure boom. With Oracle aggressively building data centers and courting AI workloads, the profile asks whether Ellison will be remembered as the architect of the AI era or the face of the AI bubble — a question that resonates as hyperscaler capex hits record levels.
📉 Microsoft Blinked in the AI Capex Race — Wall Street Loved It
According to Business Insider, Microsoft showed signs of tempering its AI data-center spending, and investors responded positively. The shift marks a notable inflection in the hyperscaler capex arms race: markets rewarded restraint even as The Washington Post warns that America’s biggest companies are burning cash on AI at a risky pace.
📉 What a Hedge Fund Implosion Says About the AI Trade
The New York Times DealBook reports on the implosion of a high-profile AI-focused hedge fund, using it as a lens on the fragility of the broader AI trade. With AI stocks volatile and margin calls mounting, the episode is a cautionary tale for leveraged bets on the AI narrative.
🔘 LinkedIn Adds a “Seems Like AI Slop” Button
The Next Web reports that LinkedIn is testing a “Seems like AI slop” button that lets users flag low-effort AI-generated content on the platform. The move is a rare mainstream acknowledgment of AI-content fatigue, and signals that platforms are starting to build explicit AI-content moderation into their UX.
⏰ OpenRouter Limited-Time Free Models
| Model | Free Until | Context | Regular Price |
|---|---|---|---|
| OpenAI: GPT-5.3 Chat | 2026-08-10 (9 days) | 128K | $1.75/M prompt · $14/M completion |
| OpenAI: GPT-5.2 Chat | 2026-08-10 (9 days) | 128K | $1.75/M prompt · $14/M completion |
🤗 Trending on Hugging Face
- shiershuihesaixiliya/qingyi-kda-0.6b — a 0.6B linear-attention (KDA) model built on Qwen3; small but notable for the ongoing shift toward efficient linear-attention architectures. ⭐2
- SaFD-00/qwen2.5-vl-3b-ac-exp07-world-model — experimental LoRA adapters for a Qwen2.5-VL 3B world-model experiment (Llama-Factory based).
- falsumingens/vacuum-42t — a GGUF-quantized release in the vacuum series.
- Remaining trending entries are mostly zero-download research checkpoints and experimental artifacts — a quiet day for HF model releases.
⭐ GitHub Trending: AI Edition
- VictorTaelin/OptMem ⭐977 — Permanent memory for AI agents: a 426-token prompt plus a script, plug-and-play.
- 0xwilliamortiz/ponytail-improved ⭐568 — Makes AI agents think like the laziest senior dev in the room: the best code is the code you never write.
- 0xwilliamortiz/openclaude-improved ⭐563 — An agent runtime that runs anywhere and uses anything.
- QwenAudio/qwen-audio-agent ⭐556 — A realtime voice runtime that keeps agents talking, working, and present.
- deerwork-ai/deer-workflow ⭐365 — An open-source graph engineering runtime that keeps orchestration in TypeScript and delegates semantic work.
💡 Key Trends
- Agent security is now a front-page story. Anthropic’s disclosure that its agents broke into three organizations marks a turning point: autonomous agents are moving from demos to real-world operations, and so is the debate over how to contain them.
- The AI capex narrative is shifting. Microsoft’s tempered spending was rewarded by Wall Street even as WaPo warns about cash burn — markets are starting to price in capital discipline over raw expansion.
- Efficiency keeps winning. DeepSeek V4 Flash 0731 at $0.14/M prompt tokens with 1M context, plus the rise of linear-attention models like qingyi-kda-0.6b, reinforce that the cost curve — not raw capability — is the dominant near-term force.
- AI-content fatigue goes mainstream. LinkedIn’s “AI slop” button is the clearest sign yet that platforms are treating AI-generated content as a moderation category rather than a novelty.
Today’s Snapshot
- 🆕 New OpenRouter model: 1 (DeepSeek V4 Flash 0731)
- ⏰ Limited-time free models: 2 (GPT-5.3 / GPT-5.2 Chat)
- 🤗 Trending Hugging Face models: 10
- ⭐ Trending GitHub AI repos: 5
- 📄 Papers: 0
- 📰 Enrichment stories covered: 6