Cybersecurity researchers at Trellix disclosed "ClawHavoc," a supply-chain attack campaign targeting OpenClaw's ClawHub skill registry with malicious packages disguised as developer, crypto, and automation tools. The malware steals crypto wallets, passwords, and developer credentials — a stark reminder that agent ecosystems with broad filesystem and messaging permissions are now priority attack surfaces.
OpenAI confirmed it has paused some frontier reinforcement-learning training to strengthen alignment, security, and monitoring standards for upcoming capability levels. Greg Brockman echoed the message, saying confidence in safety will increasingly set the pace of progress. The announcement coincided with OpenAI previewing "Private Safety Processing" — a zero-data-retention-compatible system that identifies risk patterns across related interactions without giving OpenAI personnel access to customer content.
Anthropic is preparing to give enterprise customers more control over data retention, letting them store retained data on their own cloud infrastructure rather than Anthropic's, while maintaining the 30-day retention requirement for Fable and Mythos models. The move comes as Anthropic also launched Claude Academy, a free educational platform, and rolled out Claude integrations for Gmail and Google Drive.
Trellix researchers identified a campaign using poisoned ClawHub packages to deliver malware that steals crypto wallets, passwords, and developer credentials through trusted AI agent workflows. The attack exploits the broad permissions AI agents have over files, cloud accounts, and terminal commands.
(follow-up) Sam Altman confirmed a temporary pause on frontier RL training to strengthen safety controls. OpenAI also previewed Private Safety Processing, a ZDR-compatible system that identifies risk patterns across interactions without staff accessing customer content.
Anthropic is developing a system letting enterprise customers store retained data on their own cloud infrastructure, maintaining the 30-day retention requirement while giving businesses more control. Over 100 customers including Salesforce are participating.
Nvidia rejected a report from The Information that it planned to ship a China-tailored language processing unit by year-end, stating "no China-specific LPU product" is on its roadmap. The denial underscores the ongoing flux in Nvidia's China business amid U.S. export controls.
A new Fidelity report notes AI agents settled over $73 million across roughly 176 million blockchain transactions in the year through April, while Coinbase, Stripe, and Visa develop competing machine-to-machine payment systems.
Selected posts from the edition, with links to the originals.
"Back-to-school season is here and starting today, eligible college students can get a full year of Gemini on us: US students: 1 year of Google AI Pro at no cost, 140+ countries: 1 year of Google AI"
View post on X ↗"Claude Academy is now live. Whether you're figuring out what AI is or already using Claude every day, there's a path that meets you where you are. The courses and tutorials are free and open to anyone."
View post on X ↗"We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now [paced by safety]"
View post on X ↗"Claude can now send emails in Gmail and manage files in Google Drive. Ask Claude to reply to a thread, and it drafts and sends the response. You control when it needs your approval."
View post on X ↗"We're releasing new Qwen3.8-27B GGUFs with 10% higher accuracy. Unsloth Dynamic V3 outperforms others by >10% on Div-300, KLD & more benchmarks. We also release 1-bit quants that retain 77% [accuracy]"
View post on X ↗"We will continue to offer Zero Data Retention for frontier models. As AI takes on longer, more autonomous work and delivers greater value to businesses, safety systems also need to identify risks across [interactions]"
View post on X ↗"AT&T is doing exactly what OpenAI and Anthropic fear about enterprise. they're routing 40% of employee AI usage to open models, and intend to boost it to 60-70%. coding costs down 56%"
View post on X ↗"Grok 4.6 is live on Amazon Bedrock"
View post on X ↗"GLM-5.3 API is now live. Built for coding, defensive cybersecurity, and long-horizon agentic tasks. Priced the same as GLM-5.2. Available via the official API and partner model gateways."
View post on X ↗"codex can power much more than coding tools: a tax-prep pilot processed 7,000 returns and cut preparation time by about a third. you can build your own products on our codex open-source harness"
View post on X ↗