By AZE.US Editorial Team
The White House is asking the public to trust the companies that build artificial intelligence, sell it and race to dominate its market. Now those same companies must persuade society that they can police their own competition rigorously enough to protect everyone else.
On September 29, President Donald Trump and leaders of major technology companies announced a voluntary AI-safety agreement. Trump described the document as morally binding and compared it to a constitution.
But the distance between a moral promise and an enforceable rule is the central problem. A constitution restrains power. A corporate declaration of good intentions leaves a harder question unanswered: Who can stop a company when commercial interests conflict with public safety?
Signing a document before cameras is far easier than delaying a product in which a company has already invested enormous resources.
The Auditor Also Needs Oversight
The agreement calls for internal controls and outside audits. Those are useful tools, but the existence of an audit does not establish its independence or prove that it can change a company’s decision.
The crucial questions concern who chooses the auditor, what information the auditor receives, whether unfavorable findings can be published, what happens when a developer fails to correct a danger and who can demand that deployment be suspended.
Without clear answers, an audit risks becoming evidence that a company takes safety seriously without meaningfully constraining its conduct.
The conflict of interest is obvious. A developer wants to reach the market quickly. Investors expect growth. Competitors are ready to take any opening. An employee urging another round of safety testing must challenge not only technical arguments but the commercial logic of the entire company.
A voluntary pledge does not eliminate that conflict.
Earlier Promises Produced Mixed Results
A 2025 study by Jennifer Wang and co-authors evaluated companies’ publicly disclosed compliance with eight voluntary White House commitments made in 2023. An updated version reported an average score of about 53%, while the average score for protecting model parameters from unauthorized access was 17%.
Those figures do not prove that companies broke exactly half of their promises. The researchers graded publicly available disclosures under their own methodology. But the findings expose a weakness in self-regulation: the public receives a commitment and may then have no reliable way to determine whether it was honored.
The lesson for the new agreement is direct. Safety commitments need measurable requirements, verifiable reporting and understandable consequences for failure. Otherwise, another ceremony merely refreshes the list of promises.
“Super Intelligence” as Political Packaging
Trump also directed federal agencies to replace the term artificial intelligence with “super intelligence,” according to Interfax and other reports. The White House presented the change as a way to emphasize the technology’s capabilities.
An administrative rebranding does not make a system more accurate, dependable or safe. It changes the language through which the government asks society to understand the technology.
“Super intelligence” sounds like a promise of superiority. Users may hear it as a reason to trust a machine more than its performance warrants. Government’s responsibility should include explaining a product’s limitations, not amplifying its marketing appeal.
When political presentation moves faster than verification, citizens are being sold confidence that has not yet been earned.
Why This Matters for Azerbaijan
The debate in Washington has practical consequences for Azerbaijan and other countries that use foreign AI platforms.
If these systems are introduced in banking, education, medicine or public services, the consequences of an erroneous decision arise locally, while the developer and its internal processes may be far away.
A U.S. voluntary agreement does not decide who is responsible when an Azerbaijani official relies on a faulty result. It does not tell a citizen how to challenge an automated denial. It does not replace local rules on personal-data protection or human review of decisions that affect people’s rights.
AI procurement should therefore establish who is liable for errors, who may access the data, how the system’s actions are logged and how a person can obtain meaningful reconsideration.
A vendor’s promise to act carefully does not relieve a government or business customer of its obligation to protect citizens.
Rules Must Survive a Conflict With Profit
The argument that regulation can slow innovation deserves serious consideration. Poorly written requirements can burden smaller companies and strengthen the largest incumbents.
That is a case for precise, risk-based rules. Software that helps edit text should not face the same requirements as a system influencing medical treatment or access to government services.
The real test of any safety agreement comes when compliance becomes costly and inconvenient: when a launch must be delayed, an embarrassing incident disclosed or a profitable use abandoned.
That is when society learns whether control actually exists.
The White House may welcome voluntary corporate commitments. Treating them as sufficient public protection is premature. Companies can promise responsibility. Government must ensure that responsibility survives the moment when it interferes with profit.