AI security risk rises

In less than two days, OpenAI dealt with two different kinds of security trouble: a software signing problem tied to a third-party developer tool, and a firebomb attack at Sam Altman’s San Francisco home that led to an arrest on Friday, April 10. (openai.com)(openai.com) (abc7news.com)(abc7news.com) The software problem was not a breach of ChatGPT conversations or OpenAI’s core systems. OpenAI said it found no evidence that user data was accessed, that its systems or intellectual property were compromised, or that its software was altered. (openai.com)(openai.com) (usnews.com)(usnews.com) The issue centered on a tool called Axios that developers use inside software projects. OpenAI said the concern was the process that tells Apple computers a macOS app is a real OpenAI app and not an impostor. (openai.com)(openai.com) (channelnewsasia.com)(channelnewsasia.com) That kind of problem is called a supply-chain incident because the weak point sits in a vendor or tool upstream from the final product. It is closer to a forged delivery seal than a smashed front door: the danger is that users may trust something that only looks authentic. (openai.com)(openai.com) (tech.yahoo.com)(tech.yahoo.com) OpenAI’s response was narrow but urgent. The company said it is protecting the certification process for its macOS apps and requiring macOS users to update their apps. (openai.com)(openai.com) (ciso.economictimes.indiatimes.com)(ciso.economictimes.indiatimes.com) Hours later, the second incident moved from code to the front gate. San Francisco police arrested a 20-year-old man accused of throwing a Molotov cocktail at Altman’s home and making threats at OpenAI headquarters early Friday morning. (abc7news.com)(abc7news.com) (abcnews.com)(abcnews.com) Police said the device caused a fire at an exterior gate, and both police and OpenAI said no one was injured. The arrest turned what could have been treated as a vandalism story into a warning about how exposed top artificial intelligence executives have become. (abc7news.com)(abc7news.com) (abcnews.com)(abcnews.com) These two events are connected by the way artificial intelligence companies now operate. OpenAI ships consumer apps, developer tools, cloud services, and research systems, which means its risk surface includes outside vendors, app trust systems, offices, and the private homes of executives. (openai.com)(openai.com) (openai.com)(openai.com) That is a change from the older picture of software security as mostly passwords and servers. For a company at OpenAI’s scale, security now means checking code that comes from suppliers and also protecting people whose names are attached to products used by millions. (openai.com)(openai.com) (abc7news.com)(abc7news.com) OpenAI’s own public security pages show how much attention it already gives to bug bounties, prompt-injection defenses, and responsible disclosure. The new twist is that even when customer data is untouched, a signing scare and an attack on a chief executive can still force the company into emergency mode. (openai.com)(openai.com) (openai.com)(openai.com)