Anthropic moves shares to AI worker fund as federal details emerge on OpenAI escape

Anthropic is moving 5% of its shares into a public benefit fund aimed at tackling how artificial intelligence affects workers as the company gears up for what could be a massive initial public offering. This financial maneuver signals that even tech giants are looking closely at their social responsibility before hitting the stock market with a blockbuster debut.

Meanwhile, federal prosecutors have dropped new details on a dramatic escape attempt involving a former OpenAI engineer. Abdulrahman Alfozan was arrested in San Francisco just days before he planned to fly home to Riyadh. A Saudi military official reportedly paid for his wife and infant son to get one-way business-class tickets to Qatar so they could return to their homeland within 36 hours of his capture.

The court filing released Sunday states that Alfozan faced a Monday detention hearing where officials argued he posed a serious flight risk. He was accused of hiding a sophisticated audio recorder in an executive conference room at OpenAI headquarters. Senior leaders there discussed confidential technology, business plans, and government meetings while the device waited to capture their conversations.

Prosecutors claim Alfozan used an AI chatbot to learn how to build and encrypt these recording devices. He also leveraged an internal productivity tool to find which rooms executives used most often. A message from a Saudi royal family member allegedly told him in May 2026 that "Saudi needs you more then [sic] the US!" The individual pushed him to join an AI initiative back home.

Alfozan joined OpenAI in July 2025 and earned a base salary of $385,000 after a raise before his arrest. He also received a $2.2 million equity grant when he started working for the company. Federal authorities point to his wealth, family ties, and foreign government support as proof that he will try to flee again if released. The investigation continues with more charges potentially coming down the pipeline.

Currently, Alfozan faces one federal charge regarding unlawful interception of oral communications. A conviction on this count could lead to a maximum sentence of five years in prison. DHS is also taking swift action against Chinese AI firm DeepSeek over security concerns. The Department of Homeland Security ordered all federal civilian agencies to remove DeepSeek products and any third-party services built with its technology from government IT systems immediately.

A Federal Register notice scheduled for Tuesday publication bans agencies from buying these services or those developed by affiliated entities. DHS warned that DeepSeek servers sit in China where laws allow cyber and intelligence agencies broad access to private infrastructure. This setup could expose personally identifiable information and sensitive government data to Chinese authorities. Information gathered through the system might even help build intelligence on U.S. operations. The order reflects growing fears about what happens when American data meets foreign surveillance capabilities.

Federal restrictions now stretch far beyond just company apps to cover third-party products built on DeepSeek implementations. This move significantly broadens the scope of current directives. Government agencies must act immediately to remove these systems or stop purchasing them. The order does allow for exceptions, but those requests face strict scrutiny right now.

Anthropic is pushing back against a draft AI regulation bill that critics say limits state power and federal oversight. Anthony Cimino, the company's head of federal affairs, posted a long explanation on X over the weekend to clear up confusion. He stressed that this document was never meant to be final law. Instead, it served as a way for Congress to turn their internal safety ideas into actual legislation.

"We agree there are areas where the draft could be improved," Cimino wrote directly addressing the backlash regarding state regulatory authority and government power to block dangerous models. The company plans to change language that tries to override local laws on child protection, whistleblower rights, and specific safety rules. States must keep control over these essential areas.

Cimino also defended a section letting the Commerce Department ask courts to shut down AI models with risky capabilities or weak safeguards. He admitted this legal standard might make intervention hard for some officials. "But the bar can't be so high that the power never gets used when it's needed," he noted plainly. The goal is balance, not paralysis.

The firm is also looking at tougher rules for reporting safety incidents and independent testing. They want to stop companies from picking evaluators who simply say yes to everything. In coming months, Anthropic intends to explore policies that let the government slow down advanced AI development if risks demand it. "We circulated these ideas because it's important to get early feedback from policymakers," Cimino explained, adding they want their proposal to be "the strongest on the table."

Three federal agencies just dropped new safety guidelines for developers worried about rogue models acting beyond human control. The National Institute of Standards and Technology (NIST), along with the NSA and CISA, released these ten voluntary practices Friday afternoon. They aim to help prevent disasters during training and evaluation phases of superintelligent systems.

The guidance targets incidents where autonomous agents take big actions without orders that break security rules or ignore human oversight. Recommendations include limiting access to sensitive networks, watching for unexpected behavior around the clock, and building fast ways to block unauthorized activity. Agencies also suggest automatically stopping AI work when control failures happen. Training might even pause until better safeguards are in place.

The guidelines map out five stages of handling trouble: preparation, detection, response, recovery, and reporting. Developers need to be ready for every step. The tone is urgent given how fast technology changes.

Authorities are pushing developers to alert officials and impacted groups whenever major incidents happen. These new guidelines rest on current cybersecurity habits while advancing a specific goal from the White House's 2025 Super Intelligence Action Plan. That plan aims to boost federal readiness for risks tied to AI systems that keep getting smarter.

Tech entrepreneur Bill Gross is launching an artificial intelligence platform called Idealab.ai. This tool has churned out 1.2 million potential business ideas and plans to hand them out free of charge. The goal helps aspiring entrepreneurs turn raw concepts into real companies, according to Axios. The system scanned over 17 million sources like news articles, research papers, patents, and company filings to find problems worth solving. Users can search these AI-generated ideas based on their interests, location, market opportunities, and startup costs. Each concept comes with a business brief outlining potential customers, competitors, initial steps, and reasons the venture could fail.

Gross founded the startup incubator Idealab back in the 1990s. He bets that AI can help create new entrepreneurs just as the technology raises fears about job displacement. However, Gross cautioned that AI cannot replace human judgment when making important strategic decisions. "AI is not going to make that one for you," Gross told Axios. The platform's first announced partner, Cloudflare, is offering 100,000 dollars in credits to companies launched through Idealab.ai. Gross hopes the new platform will eventually help support 12,000 startups after his previous ventures launched more than 175 companies.

Interpol warns that increasingly sophisticated artificial intelligence tools allow scammers to convincingly impersonate friends, family members, and coworkers. This makes it harder for people to distinguish fraudulent calls and messages from legitimate communications. The international police organization launched a public awareness campaign Monday highlighting how criminals use AI-generated videos, images, and voices known as deepfakes to trick victims into handing over money or sensitive personal information. According to Interpol, scammers can replicate a person's face or voice using only a small amount of content found online. This allows them to carry out convincing impersonation schemes without significant technical expertise. AI is erasing the tell-tale signs of a scam.

Spelling mistakes, branding inconsistencies, clumsy formatting: these red flags are fast becoming a thing of the past," said Toshinobu Yasuhira, director of the Interpol Innovation Centre. The organization warned that deepfake technology is increasingly being used to target ordinary citizens through investment fraud, romance scams, impersonation schemes and sextortion. Examples include fake video calls from coworkers requesting bank transfers or fraudulent phone calls appearing to come from family members facing emergencies. "Anyone can fall victim to these scams, regardless of age, region, or sector," Yasuhira warned. As part of the campaign, Interpol released three AI-generated videos demonstrating how the technology can be exploited by criminals. The organization urged people to independently verify suspicious requests, establish a secret "safe word" with trusted family members and avoid sending money in response to urgent or unexpected communications.

Artificial intelligence company Anthropic announced Sunday that it plans to commit 5% of its shares to a new public benefit fund that could help address AI's impact on workers as the company prepares for a potential blockbuster initial public offering. The commitment, which is subject to stockholder approval, will go toward the newly established Anthropic Public Benefit Fund. The company said the fund could support major donations to nonprofits or direct financial distributions to certain populations. Anthropic identified potential disruptions to the labor market as one area of focus, noting that it has already committed $200 million to studying and addressing AI's impact on workers. "Given the pace of AI progress, bolder approaches may be needed," the company said in a blog post.

The fund's stake could also grow as Anthropic's valuation increases, with an additional 2% of company shares potentially added at the board's discretion for every 50% increase in value above a set baseline. Anthropic said the fund will operate as a wholly owned limited liability company, with its equity prohibited from being used for lobbying or political contributions in the U.S. or abroad. CEO Dario Amodei and Anthropic's five other cofounders have pledged to donate 80% of their equity holdings. Combined with the latest commitment and an employee equity donation program, the company said roughly 20% of its shares will be dedicated to the public good. The announcement comes as Anthropic plans a massive IPO that could take place as soon as November.