AI security threats: Altman urges businesses to prioritise

What happened

OpenAI chief executive Sam Altman has told businesses to prioritise understanding the security threats posed by artificial intelligence, arguing it is the most useful step a company can take with AI in the near term. He made the remarks at the Oracle AI World event in Las Vegas on a Tuesday, in a conversation with Oracle co-chief executive Clay Magouyrk, as reported by The National.

Altman said that as AI use expands, the systems are continually surfacing new weaknesses. He said he was struck by the number of vulnerabilities being identified and by the scale of work required to stay ahead of them. He suggested that finding and fixing issues is the only realistic route forward, and that new methods may be needed, since patching each flaw individually across an entire supply chain may be impractical.

The National links the comments to rising adoption of OpenAI's GPT-6 model. The reporting available does not include usage figures, so the scale of that growth cannot be confirmed from this coverage.

Why it matters

The message is less about a single product than about the operating reality of enterprise AI. When a leading AI developer says that the vulnerability landscape is expanding faster than conventional patching can handle, organisations deploying AI into customer service, internal workflows and decision-making face a governance question that goes beyond IT. Security becomes part of how services are designed, tested and monitored, rather than a layer added at the end.

For leaders in digital transformation, this shifts the planning horizon. AI tools that touch customer data, claims, payments or advice can create new failure points that customers will experience directly as broken journeys, privacy concerns or lost trust. Operators will need clearer accountability for AI risk, and they may need to design for graceful failure and rapid containment, not just for accuracy and speed.

The Renascence take

Most coverage will treat this as a headline about a prominent executive urging caution. The more useful reading is that the speaker is describing a structural problem: AI systems are good at discovering weaknesses, and organisations are not yet built to absorb a steady flow of them. Trust is rarely lost in the back office. It is lost at the moment a customer meets a failure they never anticipated.

Treat AI security as a service-design problem, not only a technical one, by mapping where a failure would be felt by a customer and how quickly someone would notice. Build the journey so that a compromised tool degrades gracefully, with human fallbacks, clear communication and a visible recovery path. Behavioural research on loss aversion suggests customers forgive a disruption far more readily when they are told what happened and what is being done. Operators who wait for a perfect patch cycle will be explaining themselves to customers long before they finish fixing the bug.