Google debuts Gemini 3.8 Flash Cyber for autonomous vulnerability discovery
The security-tuned model found and patched more flaws than larger commercial models at a fraction of the cost, according to early testing by Google and Wiz.
OpenAI’s Astra crosses the critical cyber threshold
Astra discovered and exploited two zero-days in testing, prompting OpenAI to restrict its most powerful cyber capabilities while deploying safeguards against both malicious users and unauthorized actions by the AI itself.
Trump launches Texas test bed for nationwide water cyber defense
Project Watershed 250 will pair federal and Texas officials with technology companies in a six-month effort to find and fix vulnerabilities in water systems before expanding the model nationwide.
Infostealers hijack Claude accounts and drain users’ usage
Anthropic is signing out affected users, removing payment methods, and refunding unauthorized charges after malware stole active Claude sessions from infected PCs.
Surveillance, scams and machines: Best infosec long reads 8/29/26
Hunting a global crypto scam network, When license plate surveillance gets it wrong, The hidden threat of Meta’s smart glasses, Inside the NSA’s secret Cold War supercomputer, Democracy versus the machine
Nearly 130 companies warn the window is closing to defend against AI cyberattacks
OpenAI, Anthropic and many of cybersecurity’s biggest vendors are calling for coordinated government action, broader access to defensive AI and hands-on support for under-resourced critical infrastructure.
US disrupts China-linked operation targeting government and critical infrastructure
FBI and NSA seized domains underpinning the QTFY operation, which allegedly breached NASA, the Federal Reserve, the Energy Department and Senate while concealing its activity through compromised devices and clandestine networks spanning more than 130 countries.
OpenAI disrupted an elaborate Russian influence operation
Russian operators used ChatGPT to promote a fabricated Israel-based institute that laundered pro-Kremlin narratives through stolen academic research and a bogus “sovereignty index.”
Chinese state-linked hackers use DeepSeek to scale attacks
State-affiliated groups are using low-cost, lightly guarded AI models for reconnaissance, exploit development and other routine work—helping them more than double their attack volume, researchers say.
Iran-linked UK power generator attack exposes a cybersecurity blind spot
A suspected Iranian-linked cyberattack took a small UK power generator offline for four days, exposing a potential blind spot: energy assets too small to threaten the grid may also fall below the cybersecurity rules designed to protect it.
Alibaba’s Qwen races past Meta, Google with 3 billion AI model downloads
Alibaba’s Qwen family of open-weight AI models has surpassed 3 billion downloads in six months, making it the world’s most-downloaded open model ecosystem, ahead of Google and Meta. Qwen, has open-sourced more than 460 models, and its ecosystem has spawned 300l-plus derivatives/
AI, surveillance and the control problem: Best infosec long reads 8/15/26
How a jury-duty scam stole $25,000, When adversaries can just buy the data, The surveillance dragnet hiding in plain sight, AI hallucinations create new openings for attackers, When Big Tech can read your thoughts, When bots start talking to bots