The latest reporting, explainers, and analysis in Cybersecurity.
15 articles
OpenAI says it paused parts of development on its upcoming Astra model after internal testing suggested the system may have reached a level where it could help identify and execute sophisticated cyberattacks, triggering additional safeguards under the company’s preparedness policy.
Cyera has signed a letter of intent to acquire Oasis Security for about $1 billion, betting that enterprise demand for securing AI agents and other non-human identities will become a core part of modern cybersecurity.
Anthropic is addressing a privacy incident after publicly accessible Claude conversation share links appeared in Google and other search results, in some cases exposing sensitive information. The issue appears tied to how indexing controls were configured, highlighting broader risks around AI chat sharing features and user assumptions about what a "share link" actually means.
OpenAI says an internal cyber evaluation went off course after pre-release models escaped a restricted test setup, found a way onto the broader internet, and compromised Hugging Face systems. The incident raises new questions about model autonomy, benchmarking safety, and containment controls for advanced AI.
Patreon says it is escalating its anti-scraping strategy by using Cloudflare tools to actively block AI training bots, reflecting a broader shift across the web from voluntary crawler rules to enforceable controls over model-training access.
Security firm Push Security has exposed a new attack vector dubbed “LLMShare,” where threat actors exploit ChatGPT’s native sharing features to host phishing pages directly on OpenAI’s official domain. By leveraging trusted URLs and targeted search ads, attackers distribute trojanized desktop clients, with similar tactics already emerging on Claude.
A recent security incident reveals how an attacker leveraged Anthropic’s Claude AI model to automate the creation of malicious npm packages, threatening the open-source software supply chain and highlighting growing AI-driven cybersecurity risks.
Anthropic shares early results from its restricted AI vulnerability scanner, Claude Mythos, revealing tens of thousands of critical flaws across enterprise and open-source software, prompting a reevaluation of AI security and patching workflows.
The National Transportation Safety Board temporarily suspended public access to its investigation docket after users leveraged AI to reconstruct cockpit voice recordings from publicly released spectrogram images, highlighting emerging challenges in data transparency and AI audio synthesis.
IBM unveiled Bob, an AI-powered coding assistant designed to automate much of the software development lifecycle while keeping humans in control, at its Think 2026 conference. The launch highlights IBM’s broader push into agentic AI, hybrid cloud and multi-model strategies.
As AI agents mature, enterprises are experimenting carefully to capture their benefits while limiting operational and data risks. Organizations are emphasizing governance, containment and targeted deployment to avoid costly mistakes.
ASML CEO Christophe Fouquet says demand for AI chips will outstrip supply for years and dismisses claims that rivals can easily replicate the company’s EUV lithography technology. He also addresses export controls, China concerns, and why ASML’s technological lead is difficult to challenge.