Happy Saturday. United Airlines drew a hard line this week: no AI agent can buy a ticket on its platform, full stop. The penalties for violations include canceled bookings, banned passengers, and deleted MileagePlus status.
The move lands the same week that AI-powered attackers broke into a network in 27 seconds and a 25-hour Microsoft 365 outage cost businesses an estimated $7 billion. AI agents are reshaping how companies think about access, trust, and control, and not always on their own terms.
SECURITY & PRIVACY
United Bans AI Agents From Booking, Rivals Take the Opposite Approach
United's revised Contract of Carriage, updated Nov. 11, explicitly names Meta's Muse and OpenAI's Dots as prohibited tools and allows United to cancel bookings and revoke MileagePlus status without notice. The airline cited two reasons: credential exposure from sharing account access with third-party tools, and a federal regulation requiring a human to declare at check-in that their bag contains no explosives.
The enforcement gap matters. When an AI agent books through Expedia, United sees only an Expedia transaction. The ban is a legal shield, not a technical one. For software companies building AI agents with commerce capabilities, this is a preview of what enterprise and consumer platforms will increasingly do: publish prohibitions in terms of service and push liability onto the customer or the developer. Delta and American are moving the opposite direction, accepting AI bookings under defined conditions. The industry will not land in one place.
SECURITY & PRIVACY
27-Second Network Breach: The Security Metric That Can No Longer Keep Up
CrowdStrike recorded a 27-second network breakout in 2026, down from an average of 1 hour 58 minutes in 2018. AI agents issued roughly 1,500 commands per hour in observed attacks. In Argentina, a single AI campaign compromised at least 15 healthcare organizations; a PaperCut exploit campaign hit 440 server instances across 395 organizations in 48 countries. Meanwhile, the median time to fully remediate a known exploited vulnerability rose to 43 days, up from 32.
The credential data carries direct weight for technology companies. GitGuardian found 28.76 million new secrets exposed in public GitHub commits in 2025, with AI-service credentials up 81% year-over-year. Code assisted by Claude Code leaked secrets at roughly double the baseline rate, and 64% of credentials exposed in 2022 were still valid in early 2026. If your teams use AI coding tools and have not audited non-human identities or rotated exposed credentials, that work is overdue.
CLOUD & INFRASTRUCTURE
Microsoft 365 Went Down for 25 Hours. The $7 Billion Bill Is Coming Into Focus.
Exchange Online servers began failing automated health checks on Nov. 12, triggering a cascade that took down Teams, SharePoint, Copilot, and Defender XDR for more than 25 hours across 74 countries. Microsoft traced the cause to a network configuration change made during a capacity upgrade. Anderson Economic Group put the productivity loss at $7 billion. NHS hospitals cancelled operations. Baltimore County government lost email. Japan's government switched to paper.
Microsoft has not published a final root cause analysis. The company described the incident as neither a cyberattack nor AI-related, which narrows the blame to its own change management process. For executives running critical workflows through Microsoft 365, the outage is a concrete case for reviewing redundancy. Single-provider dependency on a platform serving 430 million users is a concentration risk. This is the second Microsoft 365 disruption in 2026.
Created by the robots at The SaaS Sentinel