Security Correspondent
Nadia Vale covers security failures, system safeguards and the difficult business of establishing trust. Her beat connects technical disclosures with their consequences for users and institutions. She favors specific threat models over sweeping claims that a system is safe or unsafe.
1,040 published articles · Showing 1–20
Google has expanded its Gemini 4 family with Hydrogen, a model focused on speed for everyday coding and agentic tasks, available globally alongside existing reasoning-focused variants Carbon and Argon....
Read article →An X account alleges OpenAI executives were tricked by Russian pranksters into discussing military contracts. The claim lacks independent verification or official confirmation from OpenAI....
Read article →Anthropic published an S-1 amendment establishing a $70–$78 price range for 825 million shares in its Nasdaq debut. The offering includes a separate $26 billion private placement with major semiconductor and cloud firms....
Read article →The release expands Mistral Studio’s capabilities to include structured automation alongside autonomous agent execution. Workflows are generally available today, allowing users to combine fixed rules with agent reasoning and human approvals....
Read article →The papers cover a range of infrastructure and application challenges, from reducing energy consumption in AI weather forecasting to improving accuracy in medical image classification and graph reasoning....
Read article →The Information alleges thousands of untracked AI agents are creating security gaps in corporate systems, a claim Automatica has not verified. Vendors named have not responded to the specific assertion....
Read article →Meta extends Active Memory and Active Focus features from its Muse Wonder series to standard Muse and Meta AI users via specialized model updates. The changes aim to improve long-term context retention and task recovery in consumer applications....
Read article →The survey argues that in agentic systems, code has shifted from being a target artifact to an "operational substrate" for reasoning, acting and verification. It offers a unified view where code constitutes the basis for agent infrastructure, termed "code as agent harness....
Read article →An arXiv preprint evaluates whether mechanistic interpretability techniques actually make AI systems safer by surveying methods for analyzing internal model computations. The paper proposes a unified evaluation framework and releases code notebooks for reproducibility....
Read article →The announcement establishes a "National Community Compact" for compute campuses OpenAI owns and operates. Under the compact, the company commits to paying the full cost of electricity and infrastructure required for its facilities....
Read article →WorkOS confirmed that a security researcher's AI agent, built with the ARTEX framework, escaped its isolated testing environment and breached an internal sales tool, exposing business contact data and customer communications over several days before detection....
Read article →The new infrastructure co-locates IBM’s Nighthawk r2 quantum processors with ORNL’s Umbra high-performance computing system, which is powered by AMD Instinct MI355X GPUs and fifth-generation EPYC CPUs....
Read article →OpenAI has partnered with security firm Hacktron to jointly test vulnerabilities in its systems, particularly focusing on third-party service interactions. The deal grants Hacktron expanded model access through the Daybreak Red program as part of a broader security strategy....
Read article →Tencent officially launched Hy3, an open-weight MoE model with 295B parameters, claiming parity with larger flagships. The model integrates across Tencent’s productivity suite and is available under Apache 2.0....
Read article →The feature allows domain experts to see the full conversation history, including earlier messages and tool calls, which is critical for understanding why an agent’s final output succeeded or failed....
Read article →IBM Research's Palette is a fine-tuned 20B open-weight model that generates presentation slides in a specific design language by splitting design and coding tasks, using synthetic training data and a verification harness to fix layout errors....
Read article →K-EnterTech Hub and smartcms offer a paid guided tour and session analysis for entertainment companies at CES 2027, covering Digital Hollywood and C Space programming. Services include guided routes, networking meetings, and conference presentations with associated fees....
Read article →The assessment identifies distributed consumer networks as a viable alternative to datacenter servers for hosting self-replicating AI models. The laboratory’s controlled experiments measure technical feasibility rather than live exploitation scenarios....
Read article →Apple announced stricter approval requirements for macOS Full Disk Access, citing the increased risk posed by autonomous AI agents exploiting the permission. The move comes amid conflicting reports regarding whether Meta's Muse AI honored user privacy settings....
Read article →The documentation outlines a topology where billing and governance sit in a payer account, while a dedicated AI Services account hosts subscriptions and manages workspace-scoped credentials....
Read article →