24 technology companies, including Anthropic, OpenAI, Meta, Google, Nvidia and SpaceXAI, have voluntarily committed to implement controls recommended by the White House. The commitment includes independent safety audits that will verify whether the companies’ internal controls, monitoring and detection systems actually work. Trump said the agreement is “morally binding”, but it has no legal weight and there is no new legal requirement.
What is actually happening
The central focus of the audits is on risks related to cybersecurity, biosecurity, chemical threats and unintended model actions. The audits will check whether companies’ internal controls work in practice, not just on paper. Some of the signatory companies had already committed to external audits before this agreement, so the new commitment expands an existing circle.
The agreement is entirely voluntary. There is no regulator checking compliance, and there are no sanctions for non-implementation. Nevertheless, the fact that 24 leading technology companies have signed creates a de facto international standard.
Why it matters for organisations in Israel
When an international standard emerges, even if voluntary, it affects client expectations. An Israeli organisation building an AI system for a serious enterprise client may find itself facing a demand for an external safety audit, especially if the client works with international partners or is subject to corporate governance standards.
The focus of audits on cybersecurity, biosecurity, chemical threats and unintended actions defines the areas where scrutiny is expected. A system that handles sensitive data, interfaces with critical systems or makes autonomous decisions needs to be able to prove that its controls work.
What to do this week
If you are building an AI system for an organisation, update the project document with a clear question: is the client likely to require an external safety audit? If the client works with international partners, is subject to corporate governance standards or operates in a sensitive field, the answer is yes.
Check whether your internal controls are documented and monitored in practice. An external audit checks not only what is written in documents, but what actually happens. If you do not have documentation of safety tests, logs of model decisions or a mechanism for detecting unintended actions, now is the time to build them. AI agent security requires planning in advance, not fixing after the fact.
Sources
Frequently asked questions
Is an external safety audit a legal requirement in Israel?
No, there is currently no such legal requirement in Israel. However, serious enterprise clients, especially those working with international partners or subject to corporate governance standards, may require it as part of the project.
How much does an external safety audit for an AI system cost?
The cost depends on the scope of the system, the level of risk and the depth of the audit. A basic audit can cost tens of thousands of shekels, while a comprehensive audit for a critical system can reach hundreds of thousands. It is important to budget for this in advance if the client may require it.