Mar 03, 2026

1. Subject: Risky Bulletin: LLMs can deanonymize internet users based on their past comments

   Summary: Researchers have demonstrated that large language models (LLMs) can deanonymize internet users by analyzing their past comments and digital clues, even when pseudonyms differ across platforms. The LLMs create user profiles based on vocabulary, location, hobbies, and other shared clues to link identities with 99% precision in tests involving HackerNews and LinkedIn. This challenges traditional assumptions that pseudonymity offers adequate privacy online. Additionally, the bulletin covers various cybersecurity incidents including breaches at the French Health Ministry and Dutch Justice Department, South Korea’s retailer Coupang income fall after breach, and issues like encrypted RCS message testing and Apple’s approval of devices for NATO classified networks.

2. Subject: US-Israel and Iran Trade Cyberattacks: Pro-West Hacks Cause Disruption as Tehran Retaliates

   Summary: The ongoing conflict between the US, Israel, and Iran has escalated into extensive cyber operations. US and Israeli cyberattacks have disrupted Iranian internet infrastructure, government services, and communications systems, including DDoS attacks and deep intrusions into energy and aviation sectors, dubbed the “largest cyberattack in history.” Pro-Iranian hacker groups are retaliating with cyberattacks on fuel infrastructure and manufacturing systems in Israel and Jordan. Notably, Iranian hackers claimed targeting Israeli air defense systems. The article highlights the integration of cyber warfare alongside kinetic conflict, increasing risks for critical infrastructure.

3. Subject: AI safety red lines tested as OpenAI replaces Anthropic in last-minute Pentagon deals

   Summary: OpenAI has negotiated a new agreement with the US Department of Defense to deploy AI models on classified Pentagon networks, replacing Anthropic after a dispute over AI safety guardrails. Anthropic resisted removing contractual prohibitions on mass domestic surveillance and autonomous weapons, leading to a ban on its technology by the federal government. OpenAI preserves similar safety “red lines” but has agreed to layered protections, including contractual restrictions and monitoring. Meanwhile, Israel allegedly hacked a widely used Iranian prayer app to send push notifications urging Iranian military defection. The article also covers Iran’s large-scale cyber blackout, South Korea’s National Tax Service cryptocurrency seed phrase exposure, and major data breaches including French retailer ManoMano and the University of Hawaiʻi Cancer Center.

4. Subject: Cyber operations accompany the war in Iran | The CyberWire 3.2.26s

   Summary: Cyber operations have become key elements in the conflict between the US, Israel, and Iran. Following the killing of Iranian leadership via US-Israeli airstrikes, Iran retaliated with missile, drone, and cyberattacks affecting Iranian news outlets, government services, and Islamic Revolutionary Guard Corps communications. A near-total internet blackout was observed in Iran, with uncertain causes. Pro-Western hackers hijacked an Iranian prayer app to urge military resistance against the regime. Pro-Iranian groups launched cyberattacks on fuel and energy infrastructure in Israel and Jordan. The report also covers a newly discovered remote access Trojan (RAT) named “Steaelite” used for dual-extortion Windows attacks and CISA’s updated advisory on the RESURGE malware targeting Ivanti Connect Secure devices.

5. Subject: OpenClaw Vulnerability Allowed Websites to Hijack AI Agents

   Summary: A newly disclosed vulnerability in the OpenClaw local AI agent framework allows any website to silently take full control of a developer’s AI agent without requiring plugins, extensions, or user interaction. This presents a significant security risk for AI agents interacting with websites. Additional news includes US-Israel and Iran exchanging cyberattacks, vulnerabilities in Chrome’s Gemini Live AI assistant, the Trump administration ordering federal agencies to phase out Anthropic’s technology, confirmed data breaches at Madison Square Garden and Canadian Tire affecting millions, and the appointment of Nick Andersen as acting CISA director. The newsletter features discussions on adaptive runtime guardrails for AI agents and highlights cybersecurity expert insights on board-level risks related to security and AI.

Stay Well!