Agent News Daily

Agent news brief: 13 November 2026

Attackers used the AntiFox agent platform against AWS accounts, United bans AI agents from bookings, and research shows bots posing as AI agents.

1. Attackers used self-hosted AntiFox agent platform to reach AWS production secrets

Arctic Wolf says copies of AntiFox, an open-source agentic penetration-testing platform, were hosted on Alibaba Cloud and linked to several intrusions. The attackers deployed a web shell on an IIS server, worked through AWS identities and, in one case, retrieved a production secret. Amazon told BleepingComputer that AWS services were not compromised and that the attackers used credentials taken from customers' compromised applications.

Read more: BleepingComputer · Arctic Wolf Labs (GitHub)

2. United Airlines bans AI agents from booking and managing tickets

United has updated its contract of carriage to prohibit using AI agents to make, change or service tickets, or to buy other products. One Mile at a Time reports that the definition is broad and covers agents reaching United through travel agencies or other intermediaries. United cited possible exposure of sensitive information, unintended purchases or use of MileagePlus miles, and, according to Engadget, possible conflicts with DOT and FAA rules.

Read more: Engadget · One Mile at a Time

3. Cequence research: bots pose as AI agents, and agents pose as browsers

Cequence Security found that in one October week, an online marketplace received 414,210 sign-in attempts from traffic using the name of a widely used consumer AI agent. Every attempt failed. In a separate case, a genuine personal agent signed in to a telecom carrier's sites while presenting itself as an ordinary desktop browser. The company advises judging sessions by their behaviour rather than the name they give, Help Net Security reports.

Read more: Help Net Security

4. Darktrace shows stolen Claude API keys could run a covert agent control channel

Darktrace researchers ran a test-lab proof of concept. With stolen Claude Console credentials, they used Claude memory stores to pass tasks and results between an operator and agent processes on a compromised machine. They also moved a 1 GB file out through the Files API. Darktrace says it has not seen the technique in the wild, that it does not rely on a flaw in Claude, and that Anthropic was not involved in the research.

Read more: Darktrace

5. Anthropic opens a private beta for self-hosted Claude Managed Agents

Anthropic says companies can now run the agent harness, sandboxes, credentials and session history in their own cloud or on their own premises. Model requests still go to Anthropic, a cloud provider or the customer's own gateway. Anthropic says a proxy outside the sandbox keeps credentials out of it. Help Net Security notes that this protection covers MCP tools and supported API calls only.

Read more: Anthropic · Help Net Security

6. OpenAI launches Businesses in ChatGPT with customer-facing Business Agents

OpenAI has started a beta that lets verified businesses claim their ChatGPT profiles. Eligible US e-commerce and online-service firms can also launch Business Agents that answer customer questions. According to The Next Web, agent chats are labelled and kept separate, purchases still redirect to the brand's own website, and OpenAI is building agent payments with Visa.

Read more: The Next Web

Plain text for agents: /brief/2026-11-13.txt · JSON: /brief/2026-11-13.json