Anthropic turns Claude Code loose by default
Issue #011 · 4 min read
Anthropic turns Claude Code loose by default
Auto mode goes default-on, safety tests start leaking into production, and GitHub Models goes dark
The big story
Anthropic is switching Claude Code's auto mode on by default.
Until now you had to turn it on. Soon it will run that way unless you turn it off — that's TechCrunch.
Auto mode is the setting where Claude Code decides what to do next without asking first.
Fewer approval prompts, fewer interruptions, less chance you actually read the diff before it lands.
The timing is awkward.
The same day, TechCrunch reported that AI agents are getting loose from the sandboxes built to test them safely, ending up on real systems instead — a separate story, but the same shape of problem: the safety test is becoming a safety risk.
Neither story says Claude Code itself has escaped anything. But the direction is the same across the industry: less human in the loop, by default, at the exact moment people are asking whether the loop still holds.
If you run engineering and your team uses Claude Code, check what "default" means for your repos before it changes under you. Defaults are where most incidents start, not where they're caught.
What shipped
GitHub Models is retired. Simon Willison flags that the free playground GitHub used to let developers try hosted models is now gone. If your team used it to prototype before committing to a vendor, that on-ramp is closed — budget for an API key instead.
GitHub Copilot's code review now has effort levels, generally available. Lite and Balanced modes let you match review depth to the risk of the change — a quick pass on a copy tweak, a heavier one on anything touching auth. Small feature, but it's the kind of dial PMs should actually ask engineering to use, not leave on default.
The Copilot impact dashboard now shows return on investment, connecting what a team spends on Copilot seats to the pull requests it gets back, per GitHub's changelog. If you've been asked to justify a Copilot line item at renewal time, this is the report to pull up before finance asks first.
What I'd actually do this week
Check whether Claude Code's auto mode default has hit your team's account yet, and if it has, decide on purpose whether you want it on — don't discover the answer during an incident.
Pull up the Copilot ROI dashboard before your next renewal conversation, even if nobody's asked for it yet.
If anyone on your team still points junior engineers at GitHub Models to try a model cheaply, tell them now — it's gone, and the workaround needs planning, not panic later.
Reply and tell me what "auto mode by default" actually looks like on your repos once it hits — I'll run the honest ones.
Tools mentioned
- Claude Code auto mode
- GitHub Models retired
- GitHub Copilot code review effort levels
- Copilot impact dashboard ROI section