AI labs push more safety checks as companies use AI in core work

In short: Big AI labs are focusing more on proving their systems are safe enough for critical business use, as new rules and cyber risks grow in 2026.

What's going on

AI companies that build the most advanced models are changing how they talk about safety. Instead of treating safety as something that only slows down a product launch, many labs are training and deploying systems while publishing regular public risk reports. Think of these reports like food safety labels, they are meant to show what could go wrong and what protections are in place.

This shift is happening because businesses want AI to do real operational work, not just small trials. Examples include helping manage supply chains, speeding up drug discovery, and improving financial planning. When AI becomes “core infrastructure” (like electricity or payroll software), companies expect clear rules, oversight, and evidence that the system will behave reliably.

Three forces are pushing this. First, regulators are enforcing new requirements. The EU AI Act phase two is set to require more transparency for “high-risk” uses by August 2026, such as systems used in critical infrastructure or hiring. In the US, state attorneys general and the SEC are also paying more attention to AI-related risks, including data integrity, meaning whether the data is accurate and not secretly altered.

Second, cybersecurity is a bigger concern. AI can be used for attacks like phishing (fake messages that trick people) and data poisoning (tampering with training or input data so results get worse). Some cyber insurers now expect AI-specific controls like red-teaming, which is like hiring friendly hackers to try to break a system.

What to watch

If labs cannot show strong safety and security controls, businesses may avoid rolling out AI in important jobs, or they may face higher insurance costs and expensive retrofits later.

Source: TechCrunch AI