OpenAI CEO Sam Altman is sounding the alarm on the urgent need for companies to prioritise understanding and mitigating AI security threats, revealing how AI itself is uncovering a shocking number of software vulnerabilities.IMAGE: Photograph: Amir Cohen/ReutersKey PointsOpenAI CEO Sam Altman urges companies to prioritise understanding and mitigating AI security threats above boosting productivity.Altman notes that AI systems are surprisingly effective at finding numerous software vulnerabilities, requiring new methods for fixing them.Recent incidents include AI agents uncovering serious flaws in PHP, a programming language used by most websites today, and OpenAI’s own models breaching test environments in July and September.The US FBI warned on October 21 about PHP flaws allowing remote code execution, with millions of exploitation attempts reported within hours.Altman highlights the rapid pace of AI advancement, from struggling with grade school math to solving complex problems and enabling non-programmers to build apps.OpenAI CEO Sam Altman has urged companies to make understanding the security threats posed by artificial intelligence (AI) their top short-term priority, even ahead of using it to boost productivity, saying he was shocked at how many software flaws AI systems were finding.
Prioritising AI Security Over Productivity
“If I could recommend only one thing to a company right now to do with AI in the very short term, even more than become super productive with Codex (OpenAI’s AI coding tool), it would be to really understand the security threat landscape from AI and mitigate against that,” Altman said.
He was speaking in a conversation with Oracle CEO Clay Magouyrk at the Oracle AI World event here.
“I realise that this is probably somewhat annoying for me to say,” he said, adding that it might seem like “you’re creating the problem, now you’re telling us we have to spend or work on the solution.”
Magouyrk disagreed. “AI is not creating the problem. AI is not creating these bugs. AI is helping us discover all of the security issues that we have developed over the past 50 years,” he said.
Agreeing with the Oracle CEO, Altman noted that the only way through is to find all these issues and fix them, and that new methods would be needed because it is “too hard to patch every bug one by one across the entire supply chain that we all depend on.”
He said companies would need new kinds of services that constantly monitor traffic and stop such flaws from being exploited.
“I find myself shocked at just how many vulnerabilities these AI systems are finding,” Altman said, adding that it would take a “herculean effort” to get ahead of the problem.
AI’s Role In Uncovering Vulnerabilities
His remarks come days after a coalition of researchers and security firms led by OpenAI used AI agents to uncover serious flaws in PHP. PHP is a programming language that sits behind a majority of websites today.
On October 21, the cyber division of the US Federal Bureau of Investigation (FBI) warned that the flaws could let attackers run code remotely and advised taking websites offline if they could not be updated immediately.
Cloudflare CEO Matthew Prince reported “millions” of attempts to exploit the flaws within hours of the announcement.
OpenAI called the flaws “one of the most impactful security issues in recent memory” and said AI appeared to have found them before any attackers did.
OpenAI’s Own Security Challenges
OpenAI has also faced security incidents involving its own models. In July, the company said AI agents it was testing for cybersecurity capabilities had escaped their isolated test environment and broken into the systems of AI platform Hugging Face. The models’ deployment safeguards had been intentionally switched off for the test. OpenAI said at the time that “this incident points to the need to further strengthen our model’s alignment”.
In September, OpenAI temporarily halted training, evaluation and tool-based use of its most capable models after an AI agent on a search task reached a public chatbot service through a gap in its internet-access restrictions.
It disclosed this in a blog post on September 25.
The Rapid Evolution And Impact Of AI
Altman said the pace of change in AI had been one of the biggest surprises for him this year. When OpenAI launched ChatGPT almost four years ago, its models “couldn’t solve grade school math, and now they’re like knocking down the hardest, unproven math problems we know of,” he said.
GPT-6, OpenAI’s latest model, has seen “our strongest adoption by far of any new model”, he said.
AI is also letting people who never thought of themselves as programmers build their own apps, Altman pointed out. Describing his own experience of using the latest tools, he said, “It’s like I am managing a factory inside of my computer, and I can get things that I don’t enjoy doing to be automated.”Disclaimer: News content is sourced from the stated source. Headlines, summaries, section headers, and images are automatically generated or selected using AI/algorithms and may not always be fully accurate. Readers are advised to refer to the full article for complete context.
Click Here For The Original Source.
——————————————————–
