OpenAI under scrutiny

OpenAI said on April 10 that it found a security issue tied to a third-party developer tool called Axios, and the company said it found no evidence that user data, internal systems, or intellectual property were accessed. It also said it was tightening the process used to certify that its macOS apps are legitimate OpenAI software. (openai.com) That sounds narrow, but it hits a sensitive seam in modern software: companies can lock down their own servers and still get exposed through outside code used to build, sign, or ship apps. OpenAI’s own developer security guidance tells builders to keep build tools and software dependencies patched because supply-chain risks can spread through third-party components. (openai.com) (developers.openai.com) The specific process OpenAI singled out was app certification on Apple’s macOS, which is the digital equivalent of a tamper seal on a medicine bottle. If that trust chain is in doubt, the company has to prove that downloads labeled as OpenAI software are really coming from OpenAI. (openai.com) This lands while OpenAI is pushing deeper into products that ask customers to trust it with more work, more data, and more money. Its business pages now pitch paid ChatGPT plans for Plus, Pro, Business, and Enterprise users, while its developer platform advertises security controls like single sign-on, multi-factor authentication, data residency, and role-based access controls. (openai.com 1) (openai.com 2) The menu is also getting more complicated. OpenAI’s current pricing pages show ChatGPT Business at $20 per user per month on annual billing, while the company’s public plan pages and help documents also describe flexible pricing and new Codex-only seats added to Business and Enterprise plans on April 2. (openai.com) (help.openai.com) On the developer side, OpenAI is selling more than a chatbot now. Its documentation says developers can connect models to outside services through OpenAI-maintained connectors and remote Model Context Protocol servers, which means the model can reach into tools like file stores and workplace apps if the developer allows it. (developers.openai.com) That expansion changes the political temperature around the company. When a tool mostly writes text, regulators worry about output; when a tool can connect to business systems, handle enterprise data, and trigger actions, regulators start asking who is liable when something breaks and what safeguards sit between the model and the real world. (developers.openai.com) (openai.com) OpenAI has been staffing up for that fight in Washington. Bloomberg Government reported that the company’s new in-house lobbying hires show it is taking a more campaign-style approach to influence as artificial intelligence policy battles spread across Congress, federal agencies, statehouses, and foreign capitals. (news.bgov.com) The money around those fights is growing fast. Bloomberg Government found that registered lobbying firms took in nearly $92 million in the first three quarters of 2025 from artificial-intelligence-related issues, and another report said fourth-quarter 2025 lobbying revenue tied to artificial intelligence hit a record $37.2 million. (news.bgov.com 1) (news.bgov.com 2) So the April 10 security disclosure is not just a one-day incident report. It is a reminder that OpenAI is trying to become a trusted layer for consumer apps, coding tools, workplace software, and enterprise systems at the exact moment governments are deciding how hard to police the companies building that layer. (openai.com 1) (openai.com 2)