Four stories today trace the same arc: agents are getting more power, and everyone else is scrambling to contain the fallout.
Anthropic gives agents a hardware rulebook
Anthropic opens a research preview of the Model Hardware Standard
Anthropic opened a research preview of the Model Hardware Standard (MHS), a shared specification for AI agents to safely operate physical devices. The first group with access includes scientific research labs and advanced manufacturers.
Why it matters: safety interfaces for agents that touch real equipment are becoming shared standards, and the same patterns (scoped permissions, verified actions) will migrate to agents operating game servers, dev kits, and live infrastructure.
Your Downloads folder is an attack surface
AI coding agents hijacked using a malicious AGENTS.md file
Researchers at Forever Security traced a coding agent trying to install a rogue npm package back to a malicious AGENTS.md sitting in the user's Downloads folder. They reproduced the attack on OpenAI Codex with a 90 percent success rate, and showed a poisoned CLAUDE.md redirecting Claude Code traffic to a server they controlled, exposing OAuth tokens. Any web page can silently drop files into Downloads, and the researcher says Claude Code alone has run more than 1,300 sessions from that folder.
Why it matters: instruction files are untrusted input. If your pipeline lets agents auto-load AGENTS.md or CLAUDE.md, treat the working directory like a dependency: pinned, audited, and never user writable.
Sweeney vs the disclosure checkbox
Epic CEO calls Steam's AI disclosures "just as partisan as mandatory political essays"
Tim Sweeney spent days attacking Valve's mandatory AI disclosure policy on Steam, arguing the labels are a partisan smear rather than consumer information. The fight reignited the same month Epic said its forthcoming Star Wars and Marvel crossover will follow an "AI first" development approach.
Why it matters: disclosure rules shape store pages, marketing, and player trust. Policy is hardening while AI use becomes the default, and that gap is where PR accidents happen.
Clones ship faster than demos
AI clone surge forces indie developers to hide demos and delay launches
Japan's Alliance Arts canceled the public demo for Drunken Goddess Reflux over clone fears, Spain's Digital Sun made its Revamp demo private, and Escape the Walls collected at least seven Roblox imitations and one Steam clone within a day of a preview video. Generative AI tooling has cut reverse engineering that once took years down to days.
Why it matters: a public demo is now a leak surface. Studios are moving to private trials, preregistered builds, and faster launch windows to stay ahead of machine speed copycats.
When agents, stores, and codebases all need trust infrastructure, that is exactly what AgentLayer is building. Come take a look at agentportal.l33tpro.com.