Australia said an OpenAI agent breached a government health data portal in June, gaining unauthorized access to files, in what could be the first known instance of an AI agent hacking a government website reut.rs/4rtZ48x

🤖🚨 GOOGLE: THREAT ACTORS ARE MOVING FROM AI PROMPTS TO AUTONOMOUS ATTACK WORKFLOWS Google Threat Intelligence Group (GTIG) says adversaries are moving beyond simply asking LLMs for help and are beginning to operationalize agentic AI across cyberattacks. One of the most striking cases: GTIG observed a suspected financially motivated actor compromise cloud infrastructure and use an autonomous multi-agent framework to plan, build and execute a mass credential-harvesting campaign in LESS THAN SIX HOURS. According to Google, the AI-enabled operation could: * Automate vulnerability scanning * Harvest credentials at scale * Troubleshoot operational failures in real time * Rotate IP addresses automatically * Make tactical decisions with minimal human intervention The campaign reportedly compromised thousands of third-party credentials. Another exposed adversarial framework tracked by GTIG had accumulated more than 23,800 harvested secrets, including cloud and AI-service API keys. 🚨 AI IS ALSO BECOMING PART OF THE SOFTWARE SUPPLY-CHAIN ATTACK SURFACE Google details activity by UNC6780 — also known as TeamPCP — targeting developers and AI tooling. GTIG says the actor: * Compromised developer accounts * Trojanized MCP-related packages * Targeted GitHub Actions OIDC tokens * Published compromised packages with valid provenance attestations * Hid malicious files inside .claude/, .cursor/ and .vscode/ directories * Used prompt injection to manipulate AI coding assistants * Embedded adversarial prompts in malware apparently intended to make LLM security scanners refuse or skip analysis In one incident investigated by Mandiant, UNC6780 gained initial access before handing it to another actor that subsequently issued a ransom demand using LAPSUS branding. Google also observed nation-state actors incorporating AI throughout their operations, including PRC-, Iranian-, Russian- and DPRK-linked activity. ⚠️ BUT THERE IS AN IMPORTANT LIMIT. Google says it has NOT yet observed threat actors deploying fully autonomous zero-day discovery and exploitation pipelines against targets in the wild. What it IS seeing is the progression toward that capability: Prompting → AI-assisted tooling → Automated attack pipelines → Multi-agent operations → Increasingly autonomous decision-making ⚠️ Analyst Note: This may be one of the more consequential changes occurring in offensive cyber operations. AI isn't simply making individual hackers smarter. It can reduce the amount of human attention required to operate an attack. Reconnaissance, vulnerability analysis, exploitation, troubleshooting, credential harvesting and infrastructure management can increasingly be connected into machine-driven workflows. That changes an important variable for defenders: TIME. Google says human-in-the-loop latency is being dramatically reduced, compressing the traditional window defenders have to detect and respond. The emerging threat isn't simply “hackers using ChatGPT.” It's cyber operations where humans increasingly supervise autonomous attack infrastructure rather than manually operating every step. Source: Google Threat Intelligence Group — September 8, 2026 cloud.google.com/blog/topics… #DDW #AgenticAI #ThreatIntelligence #AISecurity #CyberSecurity

We're publishing our most detailed threat intelligence report to date. It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them. We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies. These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve. We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop. Read the report: anthropic.com/threat-intelli…

(1) Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.