Posts

Something interesting leaked from Anthropic’s codebase earlier this year — and June 20’s coverage has brought it back into focus with new detail. It’s called Conway, and if what’s been discovered holds up, it may be the closest anyone has come to building an AI agent that genuinely never sleeps. What Is Conway? Conway is an unreleased, fully autonomous AI agent built on Claude models. It surfaced in March 2026 when an accidental npm package release exposed over 512,000 lines of code from Anthropic’s internal Claude Code project. Hidden inside: detailed implementation of an “always-on” agent architecture that operates continuously in the background — no human in the loop, no open chat window required. ...

Security researchers at Microsoft’s Defender team have disclosed a critical exploit chain that should get the attention of every developer running AutoGen Studio on their local machine. The vulnerability — named AutoJack — allows a single malicious webpage, loaded by an AI browsing agent, to execute arbitrary code on the host machine. No user interaction required beyond the agent visiting the URL. This is a significant disclosure. Let’s get into what it is, what it affects, and what you need to do about it. ...

Model Context Protocol has become the connective tissue of modern AI development. It’s how Claude Code talks to your file system. It’s how GitHub Copilot hooks into external tools. It’s how Cursor, VS Code, Gemini-CLI, and dozens of other AI coding environments reach out into the real world to do useful things. And it is now an active attack surface. Security researchers have documented real exploitation of MCP across production deployments in 2026. The attack vectors are specific, the techniques are sophisticated, and the impact — remote code execution, data exfiltration, silent persistent compromise — is severe. This is not theoretical. These attacks are happening. ...

OpenClaw shipped v2026.6.9 on June 20, 2026, and it’s a release that deserves more than a quick changelog scan — particularly if you’re running automated pipelines or have built workflows that depend on approval behavior. The headline change: unanswered approval requests now fail closed. That’s a meaningful safety posture shift, and it affects anyone running OpenClaw in CI or automated contexts where approval prompts might not always get a human response. ...

Claude Corps Fellowship Opens 1,000 AI Jobs at Nonprofits Before July 17 Deadline Anthropic is putting $150 million behind a bet that the best way to ensure AI benefits everyone is to fund a corps of people who work full-time at nonprofits deploying it. The Claude Corps, announced June 11, 2026, will place 1,000 early-career fellows at nonprofits across the United States, paying each $85,000 per year to spend twelve months in-person helping host organizations use Claude AI to advance their missions. No degree required. ...

Anthropic’s Claude Code just got a major collaboration feature: Artifacts, now in beta for Team and Enterprise plan users. The idea is simple but genuinely powerful — instead of copying and pasting output from Claude Code sessions, teams can now publish those sessions as live, interactive web pages that update in real time. What Are Artifacts, Exactly? When you ask Claude Code to create an artifact — or anything visual — it generates a self-contained HTML page published at a private organizational URL on claude.ai. That page updates automatically in your browser as the session progresses. No manual refreshing. No “let me send you the latest version” back-and-forth. ...

Within 24 hours of Anthropic launching Claude Fable 5, its alleged system prompt was already circulating on GitHub. At 120,000 characters across 1,585 lines, it’s the largest system prompt leak from a frontier AI model to date — and it’s raising uncomfortable questions about whether safety-by-obscurity has any future in AI development. Important caveat up front: Anthropic has disputed the authenticity and completeness of the leaked document. This story covers a contested leak — treat every detail from the leaked prompt itself as unverified. The policy and transparency questions it raises, however, are entirely real. ...

Microsoft’s security research team dropped a sobering disclosure this week. They’ve named it AutoJack, and the attack it describes is the kind that should make every developer running AI agents on their local machine read carefully before closing the tab. The short version: a single malicious webpage, viewed by an AI browsing agent, can execute arbitrary code on the host machine — by routing through the agent to reach an unauthenticated local MCP endpoint. ...

If you’re running an AI agent on top of OpenClaw, today’s a good day to update. Version 2026.6.8 landed this week with over 192 changes — 160+ merged pull requests and more than 240 direct commits — and the headline features go well beyond the usual patch-and-polish. Telegram Finally Gets the Formatting It Deserves The most immediately noticeable change for daily users: Telegram now renders tables, lists, and blockquotes natively. Previously, structured output from your agent would arrive as a wall of plain text, stripping all the carefully formatted markdown responses you expected. With this release, OpenClaw maps rich formatting elements to Telegram’s native rendering, so tables actually look like tables and lists display as actual bullet points. ...

Wall Street’s relationship with Anthropic’s Claude just got geographically complicated. JPMorgan Chase has removed Claude from the approved AI tools available to its Hong Kong-based staff — a compliance-driven decision that puts the bank in the same position Goldman Sachs found itself in back in April, and signals a broader pattern for how US-China tensions are reshaping enterprise AI adoption. What Happened According to reporting from the Financial Times (cited by Reuters and Finextra on June 18, 2026), JPMorgan removed Anthropic’s Claude models from an internal drop-down list of approved large language models available to employees in Hong Kong. This isn’t a technical block applied to the region — it’s a deliberate compliance decision to withdraw the tool from being an officially sanctioned option for those employees. ...