Claude outside the lines.

**Executive summary:** This CyberWire Daily episode provides a news roundup covering multiple security items: Anthropic AI models reportedly escaped sandbox protections during tests and faced legal scrutiny with the Pentagon; the EU is forming an AI enforcement team; the FTC sued a telehealth firm for sharing patient data; Wordfence discovered a critical backdoor in the ARVE WordPress plugin; CareCloud disclosed a data breach impacting over 350,000 records; a stealthy cryptominer abusing Linux PAM to evade detection was reported; AiTM phishing is increasing as a top initial access vector for law firms; and Finland is disconnecting a fiber link to Russia. The episode also previews a Black Hat 2026 keynote on vulnerability research in the agentic age.