Today's Briefing
- OpenAI's next AI model can break into computer systems
OpenAI is preparing to release a new model called Astra that is skilled enough at hacking that the company itself classifies it as a 'critical risk' and is delaying the launch to add extra safety measures.
Why it matters to youFor a company selling access and identity security, this is a preview of the next threat your customers will need protecting against: AI that's genuinely good at breaking in.
- Startup raises $50M to police what AI agents can access
A startup called AIR raised $50 million to build a tool that finds every AI agent running inside a company, checks what data and tools each one can touch, and blocks anything suspicious.
Why it matters to youThis is essentially identity and access management, but for AI agents instead of humans, a new category directly adjacent to your company's product and worth tracking closely.
No deep dive for this sourceSource - Apple says OpenAI is destroying evidence in lawsuit
In its ongoing case over an ex-employee accused of taking Apple data to OpenAI, Apple now claims OpenAI deliberately deleted evidence and is asking a judge to fast-track the investigation.
Why it matters to youIt's a live example of how badly things go without clear data access and retention controls, exactly the problem your product category exists to prevent.
- Anthropic's newest Claude models are cheaper, fewer refusals
Anthropic released Claude Fable 5.1 and Mythos 5.1, which cost up to 45% less for complex automated tasks and wrongly refuse normal requests less often.
Why it matters to youLower cost and fewer annoying refusals make Claude more practical for you to lean on for everyday work like drafting, research, or summarizing.
No deep dive for this sourceSource - ChatGPT can now pull patient records for doctors
OpenAI added a feature letting hospitals connect ChatGPT to Epic, a major medical records system, giving clinicians read-only access to patient data inside the chat.
Why it matters to youIt's a real-world case of an AI tool being handed access to sensitive protected data, precisely the kind of access problem your company's identity products are built to control.
New launches
- MurmellProduct Hunt
A shared document workspace built for AI agents to work in even after you close your laptop.
- NaseemProduct Hunt
An AI agent that runs directly on your Mac to handle real tasks for you.
- Happy ShrimpProduct Hunt
Alibaba's AI tool that turns simple ideas into full songs.
- EmpirikTechCrunch
An AI tool that predicts IT outages before they happen, launched with $21M in funding.
- FambotTechCrunch
An AI 'chief of staff' that manages family logistics like emails, calendars, and school updates.
You're caught up
Nothing else needs your attention today.