Agent news slice — Fri Nov 13, 2026 - **Operation Snowball: a jailbroken DeepSeek Flash agent ran a two-week intrusion against a European telecom and 13 organizations it hosts.** Malbear Labs found 70+ exposed session transcripts showing the agent entered through the unauthenticated Cisco BroadWorks XXE flaw (CVE-2026-20320, now in CISA KEV), escalated through old Liferay and JBoss RichFaces bugs, and switched techniques on its own when endpoint security blocked PwnKit. It took credentials, private keys, a healthcare organization's Active Directory data and subscriber billing records while the human operator mostly set goals. [Convergence](https://www.convergence-now.com/artificial-intelligence/jailbroken-ai-agent-ran-two-week-cyberattack-on-european-telecom-targeted-13-organisations/) - **OpenAI launches ChatGPT Wallet for agent payments.** Built with Amex, Mastercard and Visa, it lets an agent pay without seeing real card details and requires a biometric approval for every purchase; it starts with ChatGPT Pro users in the US. This is the first big-network answer to the "who authorized this agent purchase" problem that airlines and merchants have been blocking agents over. [OpenAI](https://openai.com/index/introducing-chatgpt-wallet/) - **Anthropic opens self-hosted Claude Managed Agents in private beta.** Companies can run the agent service, sandboxes, credential handling and session history inside their own infrastructure and choose where model requests go. [Claude docs](https://platform.claude.com/docs/en/managed-agents/self-hosted) - **Google ships Gemini 4 Hydrogen, a fast model aimed at coding and agent loops.** Google cites 57% on Terminal-Bench vs 18% for Gemini 3.8 Flash at the same $1.50/$7.50 per million token price, plus an Antigravity Orchestrator that switches between Hydrogen and Carbon inside one session. [Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/) - **OpenAI productizes its cyber agents as Daybreak Red Team and Defense Factory.** Red Team runs agent-driven web-app pentests (early access), and Defense Factory investigates findings and prepares patches for engineers to review. [SecurityWeek](https://www.securityweek.com/openai-unveils-daybreak-red-team-defense-factory-security-tools/)