Researchers criticize guardrails on Anthropic’s new Fable model
Anthropic has released Fable, a cybersecurity-focused AI model, but researchers report that overly aggressive guardrails are blocking even innocuous tasks.
Sections
AI & Models
524 storiesDecart launches Oasis 3 world model for real-time simulation
Decart launched Oasis 3, an interactive world model for real-time simulation, targeting autonomous vehicle developers with API access priced at $0.02 per second.
Waymo releases new benchmark model for robotaxi safety
Waymo and TU Delft released a new 'Reference Driver' model that uses active inference to benchmark autonomous driving software against human behavior in crash scenarios.
Apple brings AI-powered Siri to phones, laptops, and headsets
Apple is rolling out a revamped, AI-powered Siri across its hardware ecosystem, utilizing personal context and private cloud compute to process data while maintaining user privacy.
Anthropic releases Claude Fable 5, its first Mythos model
Anthropic has released Claude Fable 5, its first public Mythos model, which researchers say can generate complex software projects from a single prompt.
Smaller models could disrupt the economics of AI labs
Brian Armstrong predicts 80% of AI workloads will shift to 99% cheaper models within 12-18 months, a trend that could disrupt the financial outlook for major AI labs.
Anthropic launches Fable 5 as it prepares for public markets
Anthropic has launched Claude Fable 5, its first public-facing Mythos model, implementing new data retention policies and safety guardrails as it prepares to enter the public markets.
Apple unveils Siri AI with Google Gemini integration
Apple has unveiled Siri AI, a new software integration powered by Google Gemini, as it attempts to counter criticism that it is falling behind in the AI race.
Apple shifts to live-like AI demos after $250M settlement
Apple adopted "live-like" AI demos at WWDC 2026, following a $250 million settlement over false advertising claims regarding its 2024 feature announcements.
Google updates NotebookLM with Gemini 3.5 and new research tools
Google updated its NotebookLM research tool, shifting to Gemini 3.5 as the default model and adding new features to help users build knowledge bases from chat interactions.
OpenAI plans to launch a revamped ChatGPT 'super app'
OpenAI reportedly plans to launch a revamped ChatGPT 'super app' in the coming weeks, aiming to boost competitiveness and profitability ahead of a potential Initial Public Offering (IPO).
New ChatGPT Lockdown Mode targets prompt injection risks
OpenAI launched "Lockdown Mode" for ChatGPT to mitigate prompt injection risks, though the company warns the feature does not eliminate all vulnerabilities for users handling sensitive data.
Anthropic engineers reportedly assisting NSA with Mythos model
Anthropic has reportedly deployed around half-a-dozen engineers to the NSA to assist with its Mythos cybersecurity model, despite the company facing a federal supply-chain risk designation.
Mira Murati breaks silence on Thinking Machines Lab and AI governance
Former OpenAI CTO Mira Murati broke a roughly 18-month silence to detail her new startup, Thinking Machines Lab, and critique the lack of structural checks in AI governance.
Airbnb CEO Brian Chesky plans to back a new AI lab
Airbnb CEO Brian Chesky plans to back a new AI lab, though he will remain in his current role and not lead the venture himself.
Microsoft launches ASSERT for AI behavior testing
Microsoft has released ASSERT, an open-source framework that appears to help developers evaluate AI system behavior for specific product and service requirements.
Martin Scorsese joins AI startup Black Forest Labs as adviser
Martin Scorsese has joined AI startup Black Forest Labs as an adviser, though the director is reportedly using the technology solely for storyboarding purposes.
Microsoft launches Scout, an AI assistant for Microsoft 365
Microsoft is launching Scout, an AI assistant for the Microsoft 365 ecosystem built on the OpenClaw framework that requires a GitHub Copilot subscription.
Microsoft releases open source standard for AI agent control
Microsoft has released the Agent Control Specification (ACS), an open-source standard designed to help developers define and enforce governance policies for AI agents across various frameworks.
OpenAI expands Codex tools to target enterprise knowledge workers
OpenAI is expanding its Codex toolset with new plug-ins and features, targeting enterprise knowledge workers as usage grows more than three times faster than developer adoption.