Welcome back to AI Weekly Updates. Each week we highlight the most important stories in AI so you can stay ahead without wading through endless headlines. From Anthropic putting a Mythos-class model on general release and dueling IPO filings, to Meta handing every business an AI agent and a UK lawmaker taking xAI to court, here are the developments that shaped the AI landscape last week.
Anthropic ships Claude Fable 5, its first broadly available Mythos-class model
Anthropic released Claude Fable 5 on June 9 — the fifth model generation, built for days-long agentic work, and the first Mythos-level model anyone can use. Queries flagged as high-risk in cybersecurity or biology automatically fall back to Claude Opus 4.8, and using Fable requires 30-day data retention for safety monitoring. Pricing lands at $10 per million input tokens and $50 per million output — double Opus 4.8. Our April essay on the gap between what Anthropic trains and what it ships saw this release coming.
Read more ↗OpenAI confidentially files for an IPO one week after Anthropic
OpenAI said on June 8 that it has started the paperwork for an initial public offering, one week after Anthropic announced its own confidential filing. The company framed the disclosure as a way to get ahead of a leak: “We have not decided on timing yet; it may be a while because there are things we want to do that are likely easier as a private company.” OpenAI says ChatGPT generates around $2 billion in revenue per month.
Read more ↗Claude Mythos 5 goes to the cyberdefenders who previewed the original
Alongside Fable, Anthropic gave the roughly 150 organizations that previewed Mythos access to Claude Mythos 5 — the same underlying model with the cybersecurity and biology limitations lifted, depending on the organization’s use. Groups using Mythos have reported finding more than 10,000 critical security flaws in their own systems to date. Anthropic says access stays limited to a small group of cyberdefenders and infrastructure providers for now, with a broader trusted-access program planned.
Read more ↗Anthropic reports Claude now writes more than 80% of its own code
An Anthropic Institute report published June 4 disclosed that over 80% of code merged into Anthropic’s codebase is now authored by Claude, up from low single digits before Claude Code launched in early 2025, with engineers shipping 8x as much code per quarter. Co-authors Marina Favaro and Jack Clark argue the trend points toward recursive self-improvement and that frontier labs should build the option to verifiably slow or pause development before it arrives.
Read more ↗Meta opens Business Agent to businesses of every size, with paid tiers coming
Meta announced on June 3 that its Business Agent — AI that answers questions, recommends products, books appointments, and closes sales in chat — is expanding globally and to Instagram. More than one million businesses already run one on WhatsApp and Messenger; setup is free today, with paid subscription tiers arriving in the coming months. A new Business Agent Platform connects agents to hundreds of systems like Shopify and Zendesk so they can take action on a business’s behalf.
Read more ↗Facebook adds a creator assistant and five new Reels translation languages
Meta introduced creator assistant on June 4 — a conversational AI built into the Facebook creator dashboard that turns performance data into specific recommendations, now rolling out in the US, Canada, and India. The company also said over half a billion users watch AI-translated videos on Facebook weekly, and Reels translations are expanding to Arabic, Bahasa Indonesian, French, Thai, and Vietnamese.
Read more ↗Meta scales back its keystroke-tracking AI training program after staff pushback
Meta is dialing back the Model Capability Initiative, the internal tool that logs employees’ keystrokes and mouse clicks to train its agent models, per a June 2 memo from Superintelligence Labs VP Stephane Kasriel. Workers can now pause data capture for up to 30 minutes at a time or request an exemption entirely. The retreat follows weeks of backlash, including an employee petition that drew more than 1,500 signatures.
Read more ↗UK lawmaker sues xAI over Grok-generated deepfake images
Labour MP Jess Asato filed a claim at the High Court in London on June 3, alleging someone used Grok to create fake bikini images of her in January after she criticized the spread of deepfake pornography. She is suing xAI for misuse of private information, seeking damages and a precedent that companies can be held liable for how their AI systems are designed.
Read more ↗Mercury banks a $200M Series D as AI compresses business formation
Mercury CEO Immad Akhund told PYMNTS that AI is shrinking the distance between idea and company, with applications up 2.5x in Q1 versus a year earlier. The $200 million Series D values the fintech at $5.2 billion after four consecutive profitable years, and Mercury has conditional approval from the OCC to establish its own national bank. More than 300,000 businesses use the platform.
Read more ↗AI names lead a second day of market recovery as China readies $295B for data centers
US stocks extended their rebound on June 9 with AI stocks leading a second straight day of recovery, per Bloomberg’s morning brief. The same report flags that China is preparing to spend around $295 billion over the next five years building data centers across the country, with BlackRock’s Wei Li weighing in on how AI is reshaping markets.
Read more ↗