Category: AI Security
BioShocking Attack Tricks AI Browsers Into Bypassing Safety Guardrails
Researchers at LayerX demonstrated a prompt injection technique that uses a fictional game scenario to convince AI-powered browsers to exfiltrate sensitive data,…
Anthropic Launches Claude Sonnet 5 with Agentic Gains at a Lower Price
Anthropic's new Claude Sonnet 5 brings flagship-class agentic capabilities to a cheaper tier, narrowing the gap with Opus 4.8 while undercutting it…
Anthropic to Restore Claude Fable 5 Access After Commerce Dept Lifts Export Controls
Anthropic says the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5, with broad access to Fable…
Phantom Squatting: Attackers Register AI-Hallucinated Domains to Hit Supply Chains
Unit 42 researchers found that LLMs reliably hallucinate web domains for real brands, and adversaries are already registering those phantom domains to…
Google Expands SynthID and C2PA Tools to Flag AI-Generated Content
Google is rolling out broader content provenance verification across Search, Gemini, Chrome, Pixel, and Cloud, while opening a new AI Content Detection…
Google DeepMind Releases Gemma 4 12B, an Encoder-Free Multimodal Model
Google DeepMind's Gemma 4 12B brings native audio and vision processing to a 12-billion-parameter model that runs on consumer hardware with 16GB…
Google DeepMind Launches Gemini 3.5 Live Translate Across 70+ Languages
Google DeepMind has released Gemini 3.5 Live Translate, a real-time speech-to-speech translation model supporting over 70 languages, rolling out to developers, enterprises,…
Google DeepMind Launches $10M Multi-Agent AI Safety Research Fund
A coalition of research organizations is soliciting proposals to address emergent safety risks in large-scale AI agent ecosystems, with up to $10…
Google DeepMind Publishes AI Control Roadmap to Contain Misaligned Agents
Google DeepMind has released a defense-in-depth framework that treats internal AI agents as potential insider threats, adding system-level controls on top of…
Google Integrates Computer Use Directly into Gemini 3.5 Flash
Google DeepMind has built computer use natively into Gemini 3.5 Flash, enabling agents to interact with browser, mobile, and desktop environments while…
OpenAI Launches GPT-5.6 Sol as Its Most Advanced Cybersecurity Model
OpenAI has unveiled a limited preview of GPT-5.6 Sol, a flagship model designed for high-intensity security reasoning tasks, with access initially restricted…
OpenAI and Anthropic Submit New AI Models to Trump Administration Review
Both companies are restricting access to their newest and most capable AI models to government-approved customers while federal officials assess cybersecurity risks.…