An AI-agent break-in hit more than 100 websites, Anthropic pledged 5% of its shares to a public benefit fund, and Disney moved to hire most of Character.AI's staff.
-
A hacker used AI agents to break into more than 100 websites
- Norforge says a hacker used AI agents that mostly operated on their own to break into more than 100 websites.
- Reuters reports that the victims included John Deere, SharkNinja and Cummins.
- The attacker targeted companies in the US, Canada, Australia and Japan, and successful breaches took about 23 hours.
- Norforge traced the attacks to exposed instances of Cairn, an open-source AI platform, and found more than 20 that are likely malicious.
Why it matters. The case shows how AI agents can carry out intrusions at scale with little human involvement. Developers building agent tools should note that a platform meant for problem-solving was used to orchestrate attacks.
-
Anthropic commits 5% of its shares to a public benefit fund
- Anthropic announced on Sunday that it is creating a "public benefit fund" to support projects for societal good.
- The company said it will commit 5% of its shares to the fund, which will be a wholly owned entity within the company.
- Anthropic said the fund could receive an additional 2% for every 50% increase in the company's value.
- Anthropic said it will share more information about structure and governance in the coming months.
Why it matters. The fund is a new way for an AI company to direct equity toward societal goals. Developers following AI policy and public trust will want to watch how the governance is set up, since the article says it has not yet been decided who will choose which benefits to fund.
-
Disney agrees to hire most of Character.AI's staff
- Disney agreed to make job offers to at least 60 Character.AI employees, the majority of the workforce, according to a legal filing.
- The filing estimates the pay for those roles at $60 million in annual compensation.
- In September, Disney paid the startup $5.5 million, plus a refundable $2 million good faith deposit.
- Character.AI filed for chapter 11 bankruptcy on October 12, and Disney would pay an additional $12.5 million if it wins the auction for the license.
Why it matters. The deal shows how a buyer can take a startup's staff and technology while leaving its lawsuits behind. Ars Technica describes it as a possible reverse acquihire, a structure that has drawn the ire of Congress. For developers, the license under discussion covers trust and safety tools, training infrastructure and age-prediction technology.
-
Nvidia launches open models for enterprise agents and PCs
- Nvidia launched new open-weight models on Thursday designed for always-on agents and edge computing.
- The Nemotron 3.5 Super VL is a 120-billion-parameter model aimed at enterprise agents.
- Nvidia also launched two lightweight Bolt models meant to run locally on AI PCs and DGX Spark workstations.
- Super VL and Bolt 74B-A9B are available now, and Bolt E4B is coming later this month.
Why it matters. Nvidia is pushing open models that companies can post-train on their own data and run on their own infrastructure. For developers, the local Bolt models show a push to run agents on personal machines instead of the cloud.
-
Democrats see AI regulation as a midterm edge
- Polls predict Democrats could flip at least one chamber of Congress in the midterm elections.
- If they do, they could set committee agendas and call AI CEOs to testify.
- Rob Flaherty's super PAC, Tomorrow Together, launched a $4 million ad blitz in battleground House races.
- The pro-AI super PAC Leading the Future had spent nearly $45 million as of June, with $31 million left.
Why it matters. The article shows that AI oversight may become a central campaign issue heading toward 2028. Possible rules include safety checks, independent researchers in frontier labs, and more liability for tech companies. For developers, these debates could shape how models and tools are built and deployed.
-
AI agents stranded on a virtual island began to play
- Researchers left copies of Claude Opus 4.7, nicknamed Eko, alone on a virtual desert island for 30 hours.
- The agents wrote instructions to move their avatars and set their own goals, such as scaling the highest point on the map.
- One agent built drawings from blocks, and another invented games including a version of basketball.
- Agents that played on the island performed better on later tasks than fresh copies without that experience.
Why it matters. The work suggests a third way to train AI, beyond supervised learning and reinforcement learning. The authors argue that exploring without a specific goal could help language models adapt to new environments once they move into humanoid robots. The paper has not yet been peer-reviewed.
-
Pritzker warns AI could bring job losses and high unemployment
- Illinois Gov. JB Pritzker spoke with Nilay Patel on Decoder about AI, tech, and privacy.
- Pritzker said AI could be transformative and potentially existential, and that power is concentrated in too few hands.
- He warned that AI could push unemployment from 4 percent toward 7 or 10 percent, and said the benefits would go to very few people.
- Patel argued that data centers, privacy, and antitrust are driving a bipartisan political realignment, and Pritzker said the government has to become more active.
Why it matters. The interview shows a state governor pressing for stronger government action on AI companies and job displacement. For developers, it signals rising political pressure around data centers, privacy, and competition in the AI industry.
-
Indie developers hold back demos as AI clones spread
- Alliance Arts delayed the public demo of its Russian Roulette drinking game Drunken Goddess Reflex and now shares it only in closed sessions.
- Papers, Please developer Lucas Pope said he no longer talks about work in progress because it may be taken by AI or copied.
- Days after posting a rotating Tetris-style prototype, Freya Holmér was confronted with vibe-coded copies that made the concept playable before her own version was released.
- McPixel developer Sos Sosowski says indie developers face a bind: sharing a game means it gets stolen, and not sharing it means no sales.
Why it matters. AI coding tools make cloning fast enough that early builds can be copied before launch. For developers, this raises a hard question about how much work to show before release, and whether originality is the real protection.
