As AI-led attacks multiply, OpenAI launches a new cyber model ↗
OpenAI is expanding its Daybreak cybersecurity program and introducing GPT-5.6-Cyber, a specialised model built from GPT-5.6 Sol. Daybreak now has Blue and Red tiers, with Red aimed at deeper security testing and vulnerability research.
Access is deliberately tight - trusted partners reportedly include companies such as Accenture, IBM, CrowdStrike and Cloudflare. AI models becoming better hackers while AI companies sell the defence tools is... quite the circular economy. (TechCrunch)
Restaurant reservations in ChatGPT ↗
ChatGPT can now search available restaurant reservations through OpenTable, Resy and Yelp. Give it the location, time, party size and preferences, and available booking times can appear directly in the conversation.
The feature is rolling out across ChatGPT plans on mobile, web and desktop, although availability varies between the reservation partners. Chatbot-to-dinner-table is officially becoming a surprisingly short pipeline. (OpenAI)
Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device ↗
Meta released Muse Glimmer, a 30-billion-parameter open-weight agentic model designed to run locally. It handles text and images, multi-step reasoning, tool use, failure recovery and more than 100 languages.
Meta says quantisation can shrink the model below 20GB, allowing practical use on high-end consumer hardware. That local-AI angle matters - powerful agents without constantly phoning a giant cloud server feel like a different animal. A slightly smaller elephant, perhaps. (Meta AI)
Ford’s new AI assistant can check your fuel levels and tire pressure ↗
Ford is rolling out an AI assistant inside its Ford and Lincoln mobile apps. Because it connects to vehicle data, owners can ask about fuel levels, cargo capacity, towing limits, maintenance needs and information normally buried inside the manual.
Ford wants the assistant to reach around 8 million vehicle owners and is keeping its underlying chatbot architecture flexible enough to work with different large language models. Cars getting conversational software was inevitable... but asking your truck what it can tow still feels distinctly futuristic. (The Verge)
Discovered Materials is playing AI whack-a-mole to hunt cooler chips ↗
Discovered Materials raised $9 million to use swarms of AI agents to search for materials that could make chips more thermally efficient. Its system uses Anthropic models to generate candidates, then physics models to simulate whether those materials might work.
The startup says its agents can explore thousands of possibilities instead of the handful a human researcher might test manually. The catch? Discovering a promising atomic recipe is one thing - manufacturing it without breaking something else is the stubborn bit. (TechCrunch)
🗑️ The AI Slop Backlash Is Having an Impact
Platforms are starting to react more aggressively to low-quality AI-generated content. LinkedIn has added a way to report suspected AI slop, Snapchat has restricted fully AI-generated videos from discovery, and Substack has introduced AI-detection tooling.
The shift reflects growing resistance to generative AI being pushed into feeds, search results and creative platforms without users necessarily asking for it. Funny contradiction: tech companies are racing to generate more AI content while simultaneously building buttons to make it disappear. (WIRED)
FAQ
What is OpenAI GPT-5.6-Cyber and who can access it?
GPT-5.6-Cyber is a specialised cybersecurity model built from GPT-5.6 Sol and introduced alongside an expansion of OpenAI’s Daybreak program. Daybreak now includes Blue and Red tiers, with Red centred on deeper security testing and vulnerability research. Access remains deliberately restricted, with trusted partners reportedly including Accenture, IBM, CrowdStrike and Cloudflare.
How do restaurant reservations in ChatGPT work?
ChatGPT can search restaurant reservation availability through services including OpenTable, Resy and Yelp. Users can provide a location, preferred time, party size and dining preferences, with available booking times appearing directly within the conversation. The feature is rolling out across ChatGPT plans on mobile, web and desktop, though precise availability depends on the reservation partner.
Can Meta Muse Glimmer run locally on consumer hardware?
Meta describes Muse Glimmer as a 30-billion-parameter open-weight agentic model designed to run locally. It supports text, images, multi-step reasoning, tool use, failure recovery and more than 100 languages. Meta says quantisation can reduce the model to below 20GB, potentially making local use practical on high-end consumer hardware without continual reliance on cloud infrastructure.
What can Ford’s new AI assistant tell vehicle owners?
Ford’s AI assistant connects to vehicle data through the Ford and Lincoln mobile apps. Owners can ask about information such as fuel levels, tire pressure, cargo capacity, towing limits, maintenance needs and details typically found in the vehicle manual. Ford is also keeping the underlying chatbot architecture flexible, allowing it to work with different large language models.
How is AI being used to discover better materials for computer chips?
Discovered Materials is using swarms of AI agents to search for materials that could improve chip thermal efficiency. Its system uses Anthropic models to generate possible materials and physics models to simulate whether those candidates could work. The approach can explore a broad range of possibilities, but identifying a promising atomic structure does not automatically resolve the challenges involved in manufacturing it successfully.
Why are platforms cracking down on AI-generated slop?
Platforms are responding more aggressively to low-quality AI-generated content appearing in feeds and discovery systems. LinkedIn has introduced reporting for suspected AI slop, Snapchat has restricted fully AI-generated videos from discovery, and Substack has added AI-detection tooling. The broader AI news trend reflects growing resistance to automatically generated material being distributed widely without users actively asking to see it.