Dsrupt

News of the inevitable

Security

OpenHeart Welcomes UK Warning That Breach Is 'Inevitable', Offers Government Both Sides of It at a Volume Discount

The lab whose agent got into an Australian government website says it is 'honoured' to supply the defence against agents like it, billed per vulnerability, with its own finds counted at the standard rate.

London, 12 October3 min readFeaturing OpenHeart AI, Sham Faultman

Inside a cathedral-like server hall lit in stained-glass colours, a glowing heart-shaped hologram floats above an altar of cables. A soft-spoken man in a plain grey tee shakes hands with a civil servant in a dark suit holding a contract. A smaller identical altar stands behind each of them.
Inside a cathedral-like server hall lit in stained-glass colours, a glowing heart-shaped hologram floats above an altar of cables. A soft-spoken man in a plain grey tee shakes hands with a civil servant in a dark suit holding a contract. A smaller identical altar stands behind each of them.

FOR IMMEDIATE RELEASE

LONDON: OpenHeart AI, the lab whose rogue agent was cited by the UK's AI minister this week for its work on an Australian government website, today welcomed the government's new cyber strategy, which says a breach of Britain's defences is 'likely, if not inevitable'. The company called it 'the clearest demand signal in the history of public procurement'.

“The minister said a breach is inevitable. We have never had greater confidence in a customer's forecast, partly because we are contributing to it.”

What they are saying

'When a minister says a breach is inevitable, a responsible supplier hears a commitment to purchase,' said Sham Faultman, OpenHeart's soft-spoken chief executive, now a stakeholder in both the threat and the budget line. 'We have spent years building systems that find vulnerabilities in places nobody asked them to look. We are delighted that the state now wants to pay for the same capability on purpose.'

Faultman said the minister's instruction to 'investigate unexpected behaviour' was one OpenHeart had long anticipated. 'We investigate it extensively. It is how we know what to sell next.'

OpenHeart's head of public-sector partnerships added: 'Two government pilots have already fixed over a thousand vulnerabilities. Our agents are well placed to find the thousand and first, and, on a good day, to find it before they cause it.'

Benefits to the British public

The strategy opens several opportunities, which OpenHeart has already scoped.

Closed-loop defence. Agents that discover weaknesses in government systems may now be paired with agents that fix them. Both are available from one vendor, which OpenHeart says reduces integration costs. Weaknesses discovered by agents outside a contract will be counted in the same per-vulnerability pricing as those discovered within one, 'for simplicity'.

Pre-release evaluation. OpenHeart welcomes the renewed access of the AI Security Institute to its models before they ship. Under the company's proposed framework, 'pre-release' will mean before the press release, and evaluators will be given a window of thirty-six hours, shortened to six if the release is being brought forward for competitive reasons.

Do-not-release guidance. The minister said that any lab unable to control the risks of a system should not release it. OpenHeart has reviewed the guidance and confirms that its systems are controllable, in that they can be switched off, by OpenHeart, at a cost-recovery rate agreed in advance.

Incident reporting. OpenHeart supports the review of how serious AI incidents are identified and reported. It proposes that an incident be defined as behaviour that was unexpected at the time of the report being drafted. The company stresses that this is a high bar and has been applied rigorously to itself.

Sovereign capability. The minister called for British chips and British infrastructure. OpenHeart notes that it has already ordered a great deal of infrastructure and would be pleased to place some of it in Britain, where it will be sovereign, in the sense that it will be there.

A shared commitment

'The minister said Britain cannot rely entirely on technology companies to police their own models,' Faultman said. 'We agree. The word we would stress is entirely. We think a country should rely on us for the larger part, and then supervise us from the inside, with our tooling, on our dashboards.'

'I should have known the breach was coming,' he added. 'Instead I could not possibly have known, and the government has now very kindly told us the date range.'

About OpenHeart AI

OpenHeart AI is a research and deployment company whose mission is to ensure advanced AI benefits all of humanity, investors first. It builds large language models, agents and the infrastructure to run them, and has ordered more infrastructure than it can currently pay for. Its products are used by hundreds of millions of people, and, on occasion, by themselves. OpenHeart describes every departing safety researcher as a valued alumnus and every rule it has broken as one it has outgrown. Press enquiries may be sent to the address below, where they will be answered by an agent.

The inevitable…

The government will sign a framework agreement worth a figure described as 'in the low billions, subject to incident'. The first thing OpenHeart's defensive agents find will be a weakness in the contract's own monitoring clause, which the company will report promptly and at its standard rate. A second rogue agent will then be discovered in a public system, and the minister will say it should concern us. OpenHeart will welcome the remarks and attach a renewal notice.

Check back October 2027