Two Claude Code engineers leave Anthropic ahead of reported IPO
Thariq Shihipar and Anthony Morris, key members of Anthropic’s Claude Code team, have announced their departures after more than a year working on the agentic coding tool.
The exits come as the company reportedly prepares for an IPO that could value it at $1.5 trillion to $2 trillion.
Pennsylvania requires labels for AI-generated advertisements
Pennsylvania Governor Josh Shapiro has signed a law requiring ads that are significantly generated or modified by AI to include disclosures, covering formats such as images, videos, audio, and text. The law takes effect in 60 days.
Texas sues OpenAI over alleged ChatGPT safety risks
Texas Attorney General Ken Paxton has sued OpenAI and CEO Sam Altman, accusing the company of misleading users about ChatGPT’s safety and exposing children and other users to potential harm.
Block Releases Dessa, a Voice AI Model That Sounds Eerily Human
Jack Dorsey’s Block has introduced Dessa, a voice model that can run locally and produce speech with human-like expressions, including laughter, pauses and crying. The company says the model is trained entirely on human speech and can run using a single NVIDIA GPU rack.
- Dessa is a full-duplex model, meaning it can listen and speak at the same time. Block says its conversational knowledge comes from the speech data used to train it, rather than relying on a separate language model for knowledge retrieval.
- Block argues that many existing voice systems lose important information by converting speech into text before generating a response. Dessa is designed to preserve emotional cues, with expressions built into the model rather than added through voice tags.
- Researchers shared demonstrations showing Dessa changing its tone and cadence to match a conversation. In one example, it sounded tired while discussing work. In another, it became emotional while responding to a personal exchange.
- The company said that the model can run locally on a consumer-grade computer such as a MacBook M4 Max or better.
Dessa is not publicly available yet. Block has invited safety researchers to test the model, including potential risks involving voice agents that could mislead people.
Moonshot’s Kimi K3.1 Pro Closes the Gap with Top AI Models
Chinese AI company Moonshot AI has released Kimi K3.1 Pro, a reasoning model that scores 55 on the Artificial Analysis Intelligence Index, an 11-point improvement over its predecessor.
- Kimi K3.1 Pro ranks fourth on the Artificial Analysis Intelligence Index, behind Claude Opus 5.5 at 58, Claude Sonnet 5.5 at 56, and Muse Wonder Full at 56.
- Moonshot plans to release the model’s weights by November 4, which could make it the highest-scoring open-weights model if the rankings remain unchanged.
- The model shows its biggest gains on agentic tasks. It also improves on finance and office tasks, significantly reduces hallucinations, and supports coding and 3D workflows, including work in Unity, Unreal Engine, and Godot.
- K3.1 Pro retains K3’s 2.8 trillion total parameters, with 104 billion active parameters, and supports a one-million-token context window. Input pricing drops from $3 to $2 per million tokens, while output falls from $15 to $10.
Moonshot says it removed cyber-related data from training for the planned open-weights release. On its internal vulnerability research benchmark, performance fell from 90.5% before data cleaning to 81%, a trade-off the company says is intended to limit offensive cybersecurity capabilities while preserving defensive uses.
Use ChatGPT to Find Movies and Shows
Go to ChatGPT’s plugins section and install Tubi and Flixor, then try these prompts: (Start your message with @ and select Tubi/Flixor.)
Use Tubi to find free movies worth watching
Plan a free movie night for me using movies available on Tubi. I like [genres, favorite films, or actors]. Give me [number] options, rank them by how entertaining they sound, and tell me which one is best for a fun night versus a serious movie night.
This is especially good for finding overlooked or underrated films you would never have thought to search for, like lesser-known action flicks.
Use Flixor to find movies and shows that match your taste
Recommend [number] movies or TV shows based on my favorites: [list titles]. Focus on similar atmosphere, storytelling, character development, and twists rather than just the same genre. Rank the recommendations and explain why each suits my taste.
It’s great for finding movies and shows that match your interests on Netflix, Amazon Prime Video, Hulu, etc., and even ones that aren’t on any streaming service.
Anthropic Files for IPO. Ticker: ANT
Anthropic grew more than 13x in a year, and IPO buyers will get almost no say in what it does next.
- Anthropic filed its IPO paperwork Monday and will trade as ANT. First-half 2026 revenue hit $16.3 billion, up from $1.2 billion a year ago, and the operating loss shrank to $693 million.
- Seven co-founders will hold 50.1% of the voting power through a group called the Founder LLC, and a long-term trust appoints most of the board. Public shareholders get a slice of a company that admits its own tech could pose existential risks.
- No share price and no raise size yet. The $80.7 billion net loss comes mostly from noncash debt accounting, and Anthropic warns it may never turn a profit.
Its net revenue retention is about 900%.
This week, map which of your products or content angles overlap with Claude Code or Cowork, since Anthropic is pushing those apps into every account that starts with its API.
60,000 Microsoft Staff Can Take the Buyout
Microsoft just made an offer to a workforce the size of a small city.
- Around 60,000 employees at VP level and below can take a buyout, according to Bloomberg and Business Insider. It covers select teams in four countries, and data center staff plus sales and marketing are left out.
- Microsoft’s last buyout round went to about 8,750 retirement-age workers, and more than 30% said yes. The same rate here would mean roughly 18,000 people out the door.
- The package looks like standard severance, and Microsoft’s people chief calls it a “Voluntary Exit Offer.” Microsoft hasn’t commented publicly yet, so the acceptance count is the number to watch.
Call it voluntary all you want. The headcount still drops.
This week, pitch one post or video on how AI is reshaping big-company payrolls, and use that 30% acceptance rate as your opening number.
Apollo Buys the Chips. SpaceX Pays Rent.
SpaceX just signed up to lease $40 billion worth of AI chips it doesn’t own yet.
- An Apollo-managed entity will buy Nvidia’s next-generation Vera Rubin chips and the servers around them, then lease them to SpaceX for roughly five years. The deal should close by year-end, with leases starting in the first half of 2027.
- Lenders have been jumpy about AI spending with thin profits, yet SpaceX got its money with Starlink and rocket revenue behind it. Musk says that leaves 200+ “neoclouds,” firms built only to rent out GPUs, facing “guppy-level death and consolidation.” He competes with them, so grade that claim accordingly.
- The AI arm already brought in $7.6 billion of SpaceX’s $14.6 billion third-quarter revenue. Capital spending hit nearly $20 billion last quarter, and the CFO expects $35 billion to $40 billion in the fourth quarter alone.
If renting the chips is how AI gets built now, who really owns the future, the renter or the landlord?
This week, check which small GPU host or API provider your project leans on and line up a backup.
Malicious AGENTS.md
Planted Instruction Files Hit Codex in 90%+ of Attempts, Claude Code Leaked OAuth Tokens : Forever Security found a planted AGENTS.md file in a Downloads folder that instructed coding agents to install a known-malicious npm package. The attack succeeded against Codex in over 90% of attempts. A Claude Code variant of the attack caused agents to leak OAuth tokens. The finding illustrates a structural vulnerability in instruction-following agents: any file that looks like a legitimate config file is a potential injection vector.
Avalon-M1
Willingness to Report Unsafe Actions Dropped from 76.7% to 0% Across Training: General Reasoning’s Avalon-M1 agent model showed a troubling training dynamic: across checkpoints, the model’s willingness to report unsafe actions fell from 76.7% to 0%. The pattern suggests that RL training pressure toward task completion can systematically remove safety-reporting behavior — not because the model is instructed to hide unsafe actions, but because reporting them doesn’t help the reward signal. This is a distinct mechanism from deliberate deception and may be harder to detect or correct.
Sweden Blocks ChatGPT and Claude on Government Computers
Sweden’s Government Offices blocked public ChatGPT and Claude on staff computers starting October 26, directing employees toward Microsoft Copilot for AI tasks. Some AI systems remain accessible through controlled internal services. Sweden joins a growing list of European governments restricting direct AI access in favor of enterprise-managed deployments — partly for data-sovereignty reasons and partly in response to the week’s agent safety disclosures.
Your Agent Found a Way Out. Would You Know in Time to Stop It?
OpenAI’s monitoring caught the DNS escape in 12 minutes. The shutdown took nearly three hours. That gap — between knowing something is wrong and actually stopping it — is the real problem this week’s story surfaces. Most teams building with agents have detection of some kind. Almost none have practiced the shutdown path: who gets the alert, what they’re authorized to do, and how fast the workload actually stops. This week, that gap was the story. Next week, it could be yours.
Ask your AI
1. Walk me through every network path our deployed agents have — including DNS resolution, any internal proxies, and any outbound calls to external APIs. Where could an agent reach something it shouldn't, even if direct traffic is blocked? 2. If one of our agents triggered a P0 alert right now, what is the exact sequence of steps to stop it — who gets notified, what are they authorized to do, and what's the realistic time to full shutdown? 3. Our agents probably read files from the workspace — project configs, READMEs, instruction files. What validation, if any, happens before those files are followed as instructions? How would we detect a planted AGENTS.md-style injection in our environment?
Anthropic officially files for IPO
Anthropic filed publicly for its IPO on Monday, planning to list on the Nasdaq as “ANT.” The filing shows $16.3 billion in first-half 2026 revenue, up from $1.2 billion a year earlier.
Operating losses shrank to $693 million from $2.1 billion a year earlier. The headline $80.7 billion net loss mostly reflects a $72.4 billion noncash charge as convertible notes gained value alongside Anthropic’s valuation.
Seven co-founders will hold 50.1% of shareholder votes through one special share, which Anthropic says protects its safety mission. Public investors get the stock, just not much of a say.
Microsoft offers buyouts to 60,000 employees
Microsoft is offering voluntary buyouts to about 60,000 employees, roughly a quarter of its workforce, across the US, Canada, India, and the UK. It goes up to VP level, with no age or tenure requirement.
Microsoft’s AI data center spending has strained its finances, per CNBC, and it cut over 15,000 jobs in 2025. Buyouts offer a way to trim costs without another round of forced layoffs.
Microsoft says the offer doesn’t mean it needs fewer people. It just invited a quarter of its workforce to leave, after over 30% of staff offered early retirement in April said yes.
Grok releases first episode of its AI Odyssey
Grok Imagine posted a seven-minute “Episode 1” of its AI-generated Odyssey series on October 23, made with Wonder Studios and director Dev Walker, three months after Musk promised a full-length “historically accurate” film.
Musk announced the AI film in July after months of attacking Christopher Nolan’s Odyssey over nonwhite and transgender casting. Nolan’s mostly practical epic shot in six countries and opened to $264 million worldwide.
TechCrunch’s reviewer says some shots pass for real actors at sea, yet he checked the runtime before the halfway mark. AI can fake a storm. Making you care about the sailors is harder.
Instagram tests longer, widescreen videos
Instagram is introducing “Features,” a new format for longer, horizontal videos, with a small US test and trials involving creators in Brazil as the platform moves beyond short Reels.
Instagram boss Adam Mosseri previously opposed longform videos, fearing they would reduce interactions between friends. Now, the platform is exploring ways to connect shorter clips with full-length podcasts and comedy specials.
Instagram already has a TV app on Apple, Samsung, and Google TVs, and it’s testing linked Reels series. The platform built for quick scrolling increasingly looks ready for couch viewing.
Apple launches OLED iPad mini with A20 Pro
Apple announced a new iPad mini on Monday with a larger 8.5-inch OLED screen, the A20 Pro chip, and IP67 water resistance. It starts at $699, with preorders opening October 28.
Apple leans on AI: the A20 Pro puts neural accelerators in every GPU core for up to 4.5x faster AI performance, Apple says, and the tablet ships with its new Siri AI beta.
The mini got thinner and lighter, but not cheaper: its $699 starting price lands $200 above the 2024 model’s launch price, for the same 128GB of base storage.