Claude Code 'auto mode' & caps
Anthropic added an ‘auto mode’ to Claude Code so developers can grant more autonomy while cutting approval prompts — the company says it keeps safeguards but speeds workflows. Anthropic also temporarily doubled Claude usage limits as demand grows, underscoring that rationing will likely remain a feature of mainstream agent use. ( )
Available now as a research preview to Claude Teams users on March 24, 2026, auto mode will expand to Enterprise and API customers in the “coming days,” according to company and trade reporting. (macobserver.com) Anthropic’s classifier in auto mode is engineered to let routine tool calls run automatically while explicitly blocking high-risk commands—ZDNet cites mass file deletion as an example of the kinds of actions the classifier will stop. (zdnet.com) Reporting shows the classifier runs on Anthropic’s Sonnet/Opus model family (identified as Sonnet 4.6 in third-party writeups) to pre-screen each action for prompt-injection and other unexpected instructions before execution. (smartscope.blog) Anthropic explicitly positions auto mode as a middle ground between default step-by-step approvals and the --dangerously-skip-permissions flag, and the company recommends using it in controlled or sandboxed environments because it “reduces risk” but does not eliminate it. (macobserver.com) The temporary usage-promotion runs from March 13 through March 28, 2026, doubling Claude’s off-peak allotment during defined off-peak windows while leaving weekday peak-hour limits unchanged; Claude’s support page lists the promotion details and time windows. (support.claude.com) Anthropic applied the doubled off-peak allowance to Free, Pro, Max and Team plans (Enterprise excluded) and announced the move via its social account as “a small thank you,” with the change covering web, desktop, mobile and Claude Code integrations. (androidheadlines.com) Industry analysis and PYMNTS reporting frame Anthropic’s temporary doubling as part of a broader shift toward explicit rationing—providers are increasingly using daily caps, weekly reset windows and selective model downgrades to manage infrastructure costs and demand. (pymnts.com)
