📖 3 min read
Saturday roundup — five things moving in AI right now, from new models to enterprise deals to agent breakthroughs. Grab a coffee.
OpenAI rolls out GPT-5.5 with sharper safety guardrails
OpenAI’s GPT-5.5 is now the default model in ChatGPT, and the company is leaning hard on safety as the headline feature. Internal evals show a 52% improvement on harm-to-others responses and a 39% bump on suicide and self-harm handling compared with the previous default.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
Alongside the model swap, ChatGPT is rolling out Trusted Contact, an opt-in feature that lets users designate someone to be notified in serious safety situations. It’s available for personal accounts in supported regions — Business, Enterprise, and Edu workspaces are excluded for now. If you’re comparing the new tier against rivals, our AiToolCrush reviews track the major chat assistants side-by-side.
Anthropic lands PwC as a flagship Claude deployment
Anthropic announced this week that PwC is deploying Claude across its technology, deals, and enterprise reinvention practices. It’s one of the largest professional-services rollouts to date and signals how aggressively the Big Four are now embedding frontier models into client workflows.
The deal arrives on the back of Anthropic’s “Code with Claude” developer event, where the company showed off a new Claude Code feature called dreaming — agents that write notes to themselves between tasks, building up persistent memory. MIT Technology Review called it a glimpse of “coding’s future, whether you like it or not.”
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
Google DeepMind’s Co-Scientist gets published in Nature
DeepMind’s Co-Scientist, a multi-agent system built on Gemini that generates, debates, and evolves scientific hypotheses, has officially landed in Nature. The system orchestrates specialized agents that argue with each other over candidate hypotheses before surfacing the strongest ones to human researchers.
Early collaborators report it’s already accelerating work in biomedical research, with hypotheses generated by the system holding up under wet-lab validation. It’s a strong proof point that agentic architectures — not just bigger single models — are where the real gains are coming from in 2026.
OpenAI signs Malta to a national ChatGPT Plus deal
In an unusual government deal, OpenAI will give every Malta resident a year of ChatGPT Plus, contingent on completing an AI literacy course. It’s the latest in a string of national-scale agreements as OpenAI tries to make ChatGPT civic infrastructure rather than just a consumer app.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
Expect more of these. Smaller countries get a cheap path to AI fluency; OpenAI gets a built-in user base and policy goodwill. The literacy-course requirement is the interesting wrinkle — it’s the first time access has been gated on training rather than payment.
Agentic workflows quietly become the default
The pattern across this week’s news is hard to miss: every major lab is shipping multi-agent systems, not just chat models. Claude Code’s “dreaming,” DeepMind’s Co-Scientist, and a wave of OpenClaw-style autonomous workflows on indie dev Twitter all point the same direction.
For traders and bettors watching the AI sector — or anyone running prediction markets on model releases — our coverage at BetOnAI.net tracks how these launches move the needle. And if you’re picking which assistant or agent stack to actually use day-to-day, the comparisons over at AiToolCrush.com are the fastest way to get oriented.
That’s the wrap for May 23. More tomorrow.