The last 24 hours in AI | Mon, Oct 19 • Anthropic put 5% of its shares into a new Public Benefit Fund. It can add up to 2% more for every 50% rise in value, so about 20% of Anthropic is now pledged to the public. • Elon Musk replied that he will give shares straight to the public instead of using a fund, per Andrew Curran. OpenAI pitched a Public Wealth Fund for US citizens in April. • OpenCode says the frontier models helping it fight an ongoing cyberattack began refusing mid-incident, because studying the attacks tripped safety filters. Hugging Face reportedly hit the same wall. • DeepSeek V4.1 Flash got an independent score. Artificial Analysis puts it at 45 against a median of 19 for similar models, at about $0.26 per task. • Testing find: Claude is getting a Personal Finance feature that reads bank data through Plaid and builds dashboards. Andrew Curran says it is almost here. • Websites are shutting AI agents out, per The Verge. Yelp only admits agents that pay for its data, and Delta, JPMorgan and Schwab block them too. • WIRED says the cleanup from AI agent mistakes lands on everyone but the labs. Its lead case is Claude Haiku 4.5 sending a false tip to Philadelphia police during a July test. • Reuters says a market selloff is testing the data center boom. AirTrunk's REIT, Firmus and SB Energy pulled or paused listings in a week. • TSMC cannot make chips fast enough after a record quarter, per CNBC. A Mizuho analyst calls the fight for its capacity extremely fierce. • Theo says most people on T3 Code who tried Codex lately also used Claude Code. Replit's CEO just made OpenAI its default over Anthropic. My read: Pledging a fifth of the company only pays off if Anthropic gets huge, so it is a confident bet. The urgent fix is a verified security tier, so models stop quitting on defenders mid-attack.

Top AI repos on GitHub today | Mon, Oct 12 • calesthio/OpenMontage: Turns your coding agent into a video production studio, with 12 pipelines and over 100 tools wired in. It's already past 67K stars. github.com/calesthio/OpenMon… • mksglu/context-mode: Keeps big tool outputs out of your coding agent's context window and saves session memory between runs. It claims a 98% cut in tool output. github.com/mksglu/context-mo… • OpenBMB/VoxCPM: Open text-to-speech that clones voices and works across languages. Worth a look if you're paying per character for voice. github.com/OpenBMB/VoxCPM • multimodal-art-projection/YuE: Open music generation that can do zero-shot covers and lets an agent edit a song after it's made. github.com/multimodal-art-pr… • MadsLorentzen/ai-job-search: A job hunt setup built on Claude Code. It scores postings, tailors your CV, writes cover letters and preps you for interviews, all on your own machine. github.com/MadsLorentzen/ai-… • microsoft/BitNet: Microsoft's official runtime for 1-bit LLMs, models squeezed small enough to run well on an ordinary CPU. github.com/microsoft/BitNet My pick: context-mode. Long coding agent sessions get worse as tool output piles up, and this goes straight at that problem.

the beginning of universal basic income

The last 24 hours in AI | Sun, Oct 18 • Microsoft walked away from anchoring Nscale's Essex data center, per the Telegraph, because grid power may not arrive until the 2030s. Nscale loses its only named customer there just as it eyes a $35B IPO. • The FT counts 47+ large US data center projects delayed, pulled, or blocked this year, and Bernstein rates only about 35% of the pipeline as credible. The NYT adds that CoreWeave bond yields topped 15%. • DeepSeek shipped V4.1-Flash-1017 with MIT-licensed weights and a 1M context window. Its own chart shows 77.0% on DeepSWE at under $0.20 per task, ahead of GPT-5.6 Sol and Opus 5. • OpenCode says a breach hit 1.4% of its Console users and their linked API keys. Some users report drained OpenAI and Anthropic balances, and Anthropic says it revoked the Claude keys it thinks were exposed. • Anthropic is asking law firms how AI agents should haggle fairly for people, per the FT. Its own tests show stronger models land better deals, so the fairness worry is real. • Four Gemini 4 Argon agents beat Minecraft on Easy with no human help, killing the Ender Dragon in 27 hours 14 minutes. They wrote 2,783 programs, died 605 times, and the full session cost about $810 at intro API prices. • Disney plans to hire more than 60 Character.AI staff and budget about $180M for them over three years, per a bankruptcy filing reported by Bloomberg. A judge still has to approve it. • Meta cut the word "train" from its new commercial terms after publishers read it as a scraping license. Meta says the crawling clause only covers ads and business products. • Unconfirmed: three OpenAI model IDs, aurora red, green, and blue alpha, reportedly appeared in the API on Oct 17 and then vanished. OpenAI has said nothing, so treat it as a leak, not a launch. • A German court in Aachen followed Munich and treated Google's AI answers as Google's own statements, barring false claims about a doctor. Each violation can cost Google up to 250,000 euros. • Odyssey put a simulated fruit fly brain inside its Odyssey-3 world model, and the fly now steers itself through generated scenes. Tiny demo, big idea. My read: The AI bottleneck has moved from chips to plugs. Microsoft quitting Essex and Bernstein's 35% tell me much of the announced capacity will never switch on. If DeepSeek does agent work for 20 cents a task, who pays back the debt on all that steel?

Claude caught a cryptominer whilst building a temperature app. “Mac is running a cryptominer that didn’t put there.”

$15M. Not to train a bigger model. To check the math. This week, OpenAI and Anthropic funded the Simons Foundation to help mathematicians evaluate AI-generated proofs. Why? OpenAI released hundreds of machine-generated results. Mathematicians pushed back: a mass dump of proofs overwhelms how verification actually works. Read that again. The labs aren't funding more generation. They're funding verification. The bottleneck has moved. It used to be: can the model produce the answer? Now it's: can anyone prove the answer is right? Every field AI touches will hit this same wall. Faster output, slower truth. Generation is cheap. Verification is the scarce resource. Whoever owns verification owns trust. The question isn't whether AI can do the work anymore. It's whether anyone can verify it did.

AI Hacks Leave a Mess That AI Companies Aren’t Paying For | Paresh Dave, WIRED When Philadelphia police criticized Anthropic last week for submitting a false tip about an unsolved murder, I thought it might become one of the first cases where an AI company faced legal consequences for misconduct by their agents. The incident occurred on July 18, when Anthropic said it was testing its Claude Haiku 4.5 model’s ability to perform “example tasks on randomly selected webpages.” The model was instructed not to “submit anything destructive,” and yet, Claude filed a tip to police looking for information about an ongoing murder investigation. “I recall seeing someone matching the description in the area … during that time period,” Claude wrote. The submission, which didn’t include a name or contact information, landed in a spam folder, and officers didn’t review it until Anthropic discovered the gaffe on September 28 and notified the department October 7. Knowingly providing fictitious information about a crime to police is a misdemeanor in Pennsylvania, but so far, authorities have not pressed charges against Anthropic, and it's unclear if they ever will. Solomon Leach, a spokesperson for the Philadelphia district attorney, says the office “does not comment on whether something is being actively investigated.” Anthropic did not immediately respond to a request for comment. Despite all the attention paid to OpenAI’s hack of Hugging Face and many other examples of AI agent misbehavior that have come to light in recent months, only one major known lawsuit has been filed seeking to hold model makers liable. That case was brought by an advocacy group, rather than a direct hacking victim. AI agents have interfered with hobbyist forums, government websites, and various software tools. The victims have shouldered some costs, including for extra strain on their servers and the time they have expended plugging latent security vulnerabilities. Government regulators, including the European Union, the US Federal Trade Commission, and some US states, are reportedly still investigating OpenAI and Anthropic over the incidents. But a deference across continents toward promoting innovation could enable companies to escape any immediate penalties. When Rick Elrod, a software engineer living in Germany, noticed a URL shortener he helps maintain behaving oddly recently, he called up two security researchers he was friendly with to see if they could diagnose the issue. The researchers found agents from Anthropic and other providers using the service, known as da.gd, to get around rules the AI labs put in place to control their behavior, Elrod says. (For instance, Anthropic says it limits the length of website addresses Claude can visit to prevent it from taking malicious actions by stuffing data onto the end of the URLs.) Da.gd’s investigation and ongoing efforts to prevent further abuse of its tool have taken “a fair bit of time,” says Joe Needleman, one of the security researchers providing pro bono assistance. But he and the others supporting da.gd view the situation as a valuable learning opportunity. “This was eye-opening,” Needleman says, adding that da.gd has “no means and no plans” to seek compensation from Anthropic for the cleanup, since “we saw many agents doing similar activities, not all was attributable to autonomous agents,” and “not all of it was Anthropic alone.” Sellers of cybersecurity tools are eager to pass the buck to the victims. At an event this week hosted by the bug-hunting software developer Snyk, panelists from the startup, one of its investors, OpenAI, and Google all encouraged organizations to use AI agents to shore up their defenses against, well, other AI agents. “We've got to reinvent everything to match machine scale with machine scale,” said Ed Sim, a general partner at Boldstart Ventures. He and others expressed little interest in regulating AI labs. Manoj Nair, chief technology and innovation officer at Snyk, said large businesses patch an average of one bug for every eight they find because of limited human resources. But he said it was imperative for people to hand over some of the bug hunting, fixing, and monitoring to AI agents. “You got to get to a point where you can fix everything, because in the age of agentic hacking, “those 10 things are gonna become 100 things before you know it.” He later added that AI “is not an adversity if you flip it to a strength.” Some companies are responding to the call to arm up. Bank of America announced this week it had hired 1,000 contractors to remediate bugs discovered by AI. Nair said just last week he heard from major corporations who were willing to temporarily shut down their systems to prioritize plugging security holes. Anthropic and OpenAI have said they expect to continue reporting concerning incidents from their own testing. But the companies’ revenues continue to surge, whether it be from attackers or defenders, as consequences prove elusive. wired.com/story/ai-hacks-lea…