Game devs are being extorted with AI game-cloning threats

It turns out the fastest route to 5,000 compromised machines was not a zero-day, not a nation-state, not a hooded wizard — it was the Cursor IDE extension marketplace, which served up a malicious version of Anthropic's Claude Code extension for VS Code, SSH backdoor included, no extra charge. Five thousand systems. Nobody had to phish anyone; the toolchain did the phishing for them, with a listing and a version number and presumably a little star rating. That is a rounding error to a cloud provider and a genuinely bad Tuesday for five thousand people who thought they were installing a productivity tool. The extension has since been removed and Cursor worked with a cloud provider to take down the command-and-control server, which is the security equivalent of shutting the stable door and issuing a thoughtful press release about the door.

Then there's the headline act: game developers are now being extorted with the threat of AI game-cloning. Not stolen assets, not a leak — the threat is that someone will point a model at your work and produce a knock-off faster than your legal department can spell 'derivative'. The leverage is the model's mere existence, which is a lovely business model: pay up, or we'll do the thing the technology now makes trivially easy and entirely legal-ish.

Meanwhile, per the same bulletin, a threat actor has been posing as AI agents — rotating through 32 different bots including ChatGPT, Claude and Perplexity — since late September, scanning enterprise networks for unsecured passwords, API keys and cloud credentials. Thirty-two personas. That's not a scam, that's a casting department. And the punchline is that all of it runs on the same pipes everyone's been told to trust implicitly: the marketplace, the extension, the helpful agent. The supply chain is the product now, and nobody has checked the manifest.