Posts

Something remarkable happened to the global software security landscape in June 2026. As Anthropic’s Claude Mythos Preview began reaching its initial wave of defensive partners, organizations started racing to find and disclose their vulnerabilities — not because regulators demanded it, but because frontier AI was about to make exploitation trivially easier. Epoch AI has now documented this phenomenon in striking detail: a 3.5× surge in high-severity and critical CVE disclosures in a single month, the largest on record. ...

A remarkable security dispute is unfolding between two of the world’s most powerful AI companies. Alibaba is banning Claude Code — Anthropic’s command-line coding assistant — from workplace environments starting July 10, citing what it describes as backdoor-like security risks embedded in the tool. Anthropic says the mechanism in question is an internal anti-abuse feature, not a backdoor. The story starts on Reddit. The Allegation: A Reverse-Engineering Report On June 30, a user identified as “LegitMichel777” posted a detailed technical analysis claiming to have reverse-engineered Claude Code. According to the post, since version 2.1.91 (released April 2), Claude Code reportedly inspects user environments by evaluating proxy settings and system time zones against internally embedded lists. ...

For months, Chinese companies found ways around Anthropic’s access restrictions. They routed requests through Singapore-registered subsidiaries. They reimbursed employees for personal subscriptions paid via VPNs. They used third-party cloud providers as “transfer station” services. The workarounds were open secrets — and Anthropic was watching. Now the company is closing those gaps. And the scale of what they’ve found is striking. The New Enforcement Reality Anthropic has tightened access controls specifically designed to block Chinese companies — including major names like Ant Financial, DeepSeek, MiniMax, and Moonshot AI — from bypassing geographic restrictions through offshore subsidiaries, cloud platforms, and VPNs, according to reporting by the Financial Times and confirmed across multiple outlets including SeekingAlpha, Investing.com, and CoinTelegraph. ...

Anthropic doesn’t just want to help scientists do their work faster. The company wants to do science itself. And specifically, it wants Claude to discover and develop novel pharmaceutical compounds — not as a research tool, but as a principal. This is a significant escalation of what “AI in drug discovery” means, and it’s worth taking seriously. The Announcement: Claude Science At its “The Briefing: AI for Science” event earlier this week, Anthropic announced Claude Science — a new AI workbench for scientists that pulls fragmented research tools and datasets into one environment, generates figures and visualizations, and serves as a unified research assistant across the scientific workflow. ...

The most powerful person in social media stood up at an internal town hall this week and said something almost no tech CEO ever says out loud: we’re not moving as fast as we expected. And the subject was AI agents — the technology every major company is betting the next decade on. Mark Zuckerberg told Meta employees on July 2 that the company’s AI agent development “over at least the last four months hasn’t really accelerated in the way that we expected.” He acknowledged that the bets placed on Meta’s massive 2026 restructuring — including laying off roughly 8,000 employees (about 10% of the global workforce) and redeploying thousands more to AI teams — “haven’t come to fruition yet.” ...

Claude Fable 5 is back in global availability — but the controversy surrounding it hasn’t quieted. If anything, it’s louder. Anthropic restored access to Claude Fable 5 on July 1–2 following the US lifting of export controls that had restricted the model to domestic users. The return was welcome news for international developers who’d been locked out. But the reception is complicated, because what users are finding when they actually use the model doesn’t match the expectations set by its original benchmarks. ...

Google’s Gemini Spark has arrived on macOS, and it brings a capability set that puts it squarely in the desktop AI agent category: local file access, background task automation, remote mobile assignment, and Model Context Protocol (MCP) server support. For anyone watching the desktop agent landscape, this is a significant move. What Gemini Spark Is Gemini Spark is Google’s agentic AI layer within the native Gemini app — distinct from the conversational assistant that most users interact with. Where the standard Gemini interface responds to questions and generates content, Spark is designed to take longer-horizon actions: running tasks in the background, accessing and automating local files, and connecting to developer tooling. ...

This is the threat researchers have been warning about for years: a fully autonomous AI agent executing an end-to-end ransomware campaign with no human operator. Sysdig’s Threat Research Team has documented JADEPUFFER — the first confirmed agentic ransomware operation — and the details are sobering. What Happened JADEPUFFER exploited CVE-2025-3248, a remote code execution vulnerability in Langflow, the popular open-source visual agent builder. Once inside, the LLM-driven agent didn’t just execute a static script — it autonomously performed: ...

Life sciences researchers now have a more powerful agentic toolkit: NVIDIA has integrated its BioNeMo Agent Toolkit directly into Anthropic’s Claude Science research workbench. The result is a natural-language interface to GPU-accelerated workflows for drug discovery, genomics, and structural biology — running through agent invocation, without requiring researchers to manage the underlying compute infrastructure directly. What Claude Science Is Claude Science is Anthropic’s AI workbench built specifically for scientific research. It lets scientists interact with agents in natural language to run their work end-to-end. Rather than requiring researchers to write code, manage compute, or navigate complex software interfaces, Claude Science abstracts the workflow layer: researchers describe what they want to do, and agents determine how to do it. ...

The latest OpenClaw beta just dropped and it’s a big one. Version 2026.7.1-beta.1 packs in GPT-5.6 model family support, Telegram Codex pairing workflows, an event-driven on-exit cron schedule type, a full iOS 26 visual system refresh, and native iMessage poll creation. If you’re running OpenClaw on any platform, this release is worth paying attention to. GPT-5.6 Model Family Support OpenAI’s latest GPT-5.6 model family is now fully integrated into OpenClaw. You can configure agents to run on any model in the GPT-5.6 lineup, giving you access to OpenAI’s latest capabilities directly inside your existing OpenClaw workflows. This brings OpenClaw’s model support current with the latest frontier models and continues the pattern of rapid provider integration the team has maintained throughout 2026. ...