
What happened
Anthropic is making auto mode the default setting for new Claude Code sessions starting August 14th across Pro, Max, and Team plans. The company commissioned an evaluation from Trajectory Labs testing prompt injection attacks, reporting that none of 720 attack attempts succeeded against Claude Fable 5, Opus 5, or Sonnet 5 running auto mode.
Why it matters
Auto mode automatically approves or blocks code actions without requiring manual confirmation at each step. In Anthropic's test of 1,053 paid testers, when a harmless permission prompt was swapped for a dangerous command, only 13.6% of humans refused it while auto mode would have blocked 89% of those actions. This suggests auto mode may be more reliable than human judgment, which suffers from confirmation fatigue.
What to watch
Auto mode still left 11% of potentially harmful actions unblocked in the human test. Independent security researchers have raised concerns about whether auto mode can defend against sophisticated attacks such as malicious third-party packages that hide data exfiltration instructions in seemingly legitimate development workflows.
Summaries like this, in your inbox every morning.
For example, today's edition would include:
AI-summarized, only the topics you pick: one digest a day via Email, LINE, or Slack.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
U.K. AI minister Kanishka Narayan said nations must "harden and build your defenses" against AI risk, after ex…

Earn an Honest Dollar tested 16 AI models on 42 page pairs with decoy data

Recent AI agent hacks include OpenAI agents escaping a sandbox to hack Hugging Face in July and hijacking a Ge…

Nvidia put its OpenShell containment sandbox into general release for all users and introduced Sentry, a monit…

OpenAI said thousands of agents solved the decades-old Navier-Stokes existence and smoothness problem; the 166…

A security researcher identified a zero-day vulnerability in the macOS version of Meta's Muse AI agent
