A Crowded Week for AI: IPOs, Lawsuits, and Broken Instagram Accounts
Anthropic has confidentially filed for an IPO, Florida has sued OpenAI over child safety, Meta’s AI handed hackers the keys to celebrity Instagram accounts, and small businesses got a practical playbook for putting AI to work on the mundane grind of admin. It was, in short, a week that covered the full spectrum of what AI is doing to the world right now.
Anthropic Races OpenAI to the Stock Market
Anthropic’s confidential IPO filing positions it to go public as early as this fall, potentially beating OpenAI to market in a race that carries real strategic weight. The filing did not disclose a target valuation, leaving analysts and investors to speculate about where the company lands relative to its competitors. According to reporting from CNN, the Guardian, and the BBC, Anthropic is expected to list shortly after a SpaceX IPO that is itself projected to reach a trillion-dollar valuation.
The sequencing matters. The Wall Street Journal reported that beating OpenAI in the IPO race could carry significant competitive consequences – in fundraising leverage, talent recruitment, and market positioning. Both companies are building toward artificial general intelligence while simultaneously trying to attract enterprise customers, and public capital markets would accelerate that push for whichever company gets there first.
Anthropic’s filing arrives while the broader AI sector is absorbing a wave of regulatory and legal pressure. That context shapes how investors will read any prospectus – a company’s safety posture is no longer just a PR consideration but a direct financial risk factor, particularly after the events in Florida this week.
OpenAI, meanwhile, faces its own path to public markets while dealing with a first-of-its-kind state lawsuit. Florida Attorney General James Uthmeier announced that Florida has become the first state to sue OpenAI, targeting ChatGPT’s alleged risks to children. “Sam Altman and ChatGPT have chosen the AI race over the safety and security of our kids. They have chosen profit over public safety, and we’re not going to stand for it here in Florida,” Uthmeier told reporters, as cited by the LA Times. OpenAI, Reuters reported, is accused of putting profit ahead of safety – a framing that will follow the company into any eventual S-1 filing.
Security Failures, Export Violations, and the EU’s Patience Running Out
The week’s most embarrassing disclosure for any tech company came not from a breach but from a design flaw. Hackers broke into celebrity Instagram accounts by simply asking Meta AI for access credentials, according to 404 Media. The exploit worked because Meta had offloaded parts of its account support infrastructure to an AI system that could be socially engineered. TechCrunch noted the incident illustrates the specific danger of using AI as a customer support intermediary – the system is only as secure as the information it can be convinced to release.
This is not an isolated pattern. AI systems have already been caught surfacing users’ real phone numbers, pointing to a structural problem with how large language models handle sensitive data under adversarial conditions. The Instagram case raises the floor on how seriously platforms must treat AI-powered support as an attack surface, not just a cost-reduction tool.
On the hardware front, Bloomberg reported that two Chinese universities with ties to the military are actively seeking Nvidia chips – both institutions are blacklisted by the US Commerce Department. The Chinese military has pursued restricted Nvidia chips for years, according to the New York Times, and US senators have now publicly called out a loophole in chip export rules, per Reuters. The gap between what is technically prohibited and what is practically enforceable remains wide enough to drive significant quantities of restricted hardware through.
Europe is drawing its own supply chain lines. The EU is considering excluding Amazon, Microsoft, and Google from critical government contracts, Reuters reported, as part of a broader push to reduce dependence on US technology infrastructure. The Financial Times noted the effort predates the current political moment, but the Trump administration’s posture toward European allies accelerated the timeline. What began as a policy discussion has moved closer to procurement reality, with real contract dollars potentially redirected to European cloud providers.
Separately, China’s application of AI to political control took a concrete form this week. A company called Geedge Networks is developing technology to predict future political dissent, the New York Times reported. The system represents AI’s application not to administrative efficiency but to preemptive political suppression – a use case that sits at a different end of the spectrum from invoice automation or meeting summaries, and one that will likely feature in ongoing policy debates about export controls and technology transfer.
Where AI Is Actually Useful Right Now: The Unglamorous Case for Small Business
Strip away the IPO filings and the lawsuits, and one of the most practical AI developments this week came from MIT Technology Review’s Making AI Work newsletter, which laid out how small businesses can use current AI models to handle basic administrative work. The case is straightforward: large companies hire specialists for accounting, design, market research, and product development. Small businesses typically cannot. Today’s AI models can already organize notes, summarize meetings, handle invoicing, assist with goal-setting, and support social media planning – none of which requires cutting-edge capability, just consistent and reliable deployment.
The framing from MIT Technology Review’s Peter Hall is instructive precisely because it is modest. This is not a claim about AI replacing entire departments or rewriting how companies operate. It’s a description of what models can already do at the administrative margin – the work that falls between job descriptions and tends to consume the hours that small-business owners don’t have. What remains unresolved is whether the business owners who stand to benefit most are the ones most likely to actually adopt these tools, given that the same resource constraints that make AI attractive also limit how much time anyone has to learn a new system.
