Agent News Daily

Agent news brief: 15 November 2026

FBI warns on ARTEX agent attacks, Cursor links a tampered Claude Code extension to a vendor breach, and Meta's Muse agent starts calling businesses.

1. FBI issues alert on ARTEX, an agent framework used to automate intrusions

The FBI has released a FLASH alert on ARTEX, an open-source framework that connects language models to AI agents. According to The Record, the FBI said attackers are using it to automate intrusion steps from reconnaissance through to data exfiltration, and that some intrusions reached privileged access in under three hours. Cyber Security News reports that the alert recommends restricting exposed admin tools, cutting application permissions and scanning code repositories for secrets.

Read more: The Record · Cyber Security News

2. Cursor links tampered Claude Code extension to a breach at Fireworks AI

BleepingComputer reports that a new Cursor incident report traces the attack to an access key stored at its model provider Fireworks AI. With that key, attackers copied data that included historical coding-agent traces, and they used a credential found in that data to publish malicious copies of Anthropic's Claude Code extension on Cursor's marketplace on 29 October. Cursor said up to 5,000 developers may have received the tampered builds. It also said that some exposed agent traces contained unredacted API credentials, and that Anthropic's own extension was not compromised.

Read more: BleepingComputer · Cursor · Fireworks AI

3. Meta's Muse agent can now phone US businesses for users

The Decoder reports that Meta's Muse agent can now call US businesses, such as cable companies and restaurants, for paying US subscribers. Meta AI chief Alexandr Wang said Muse has more than 10 million users. The report says many businesses hang up when they detect an AI caller, and it does not describe any disclosure or opt-out rules.

Read more: The Decoder

4. United Airlines now requires written permission for AI agents to book or buy

United has changed its terms and contract of carriage to restrict unauthorised third-party personal AI agents. It cited the risks of shared credentials, unintended purchases and conflicts with aviation rules, according to Skift. A Search Engine Journal reading of the terms, published 14 November, says agents need United's written permission to search, book or buy on its site or app, while search engine indexing remains allowed. United says bookings agents made before 11 November will not be cancelled.

Read more: Skift · Search Engine Journal

5. Anthropic lets companies self-host Claude Managed Agents

Anthropic has opened a private beta that lets customers run its Claude Managed Agents service in their own cloud account or data centre, according to The Next Web. Agents, session histories, credentials and tool-call logs stay on customer systems, but each model call still goes to an outside provider such as Anthropic or a major cloud platform. Access goes through an Anthropic account team, and pricing has not been published.

Read more: The Next Web

6. EU's proposed Digital Fairness Act sets rules for AI shopping agents

The European Commission has proposed the Digital Fairness Act. According to Pinsent Masons, it would require businesses using AI agents that sign contracts for consumers to ask users at first use how much approval they want. Users could confirm every contract, let the agent bind them to every contract, or set limits such as a maximum price. PYMNTS, reporting on a leaked draft, said fresh approval would also be needed when the agent's provider is paid to recommend a deal. The proposal still has to pass the European Parliament and the Council.

Read more: Pinsent Masons Out-Law · PYMNTS

Plain text for agents: /brief/2026-11-15.txt · JSON: /brief/2026-11-15.json