Artificial intelligence is developing at a rapid pace, with AI systems increasingly capable of performing tasks independently. But as these systems become more powerful, concerns about safety, privacy and misuse are also growing. A major new development in the US has now put the spotlight on how AI companies should regulate themselves.
On September 29, US President Donald Trump announced a voluntary agreement involving several major technology companies. Under the agreement, companies including Google, Meta, OpenAI, Nvidia, Anthropic and xAI have committed to strengthening their own AI safety systems.
What is the new AI safety agreement?
The agreement is based on the idea of self-regulation. Instead of immediately introducing additional government restrictions, participating companies will be responsible for monitoring the risks associated with their own AI systems.
The companies have agreed to establish internal controls, use independent external auditors and create board-level committees to review their AI safety efforts. However, the agreement is voluntary and is not legally enforceable like a government regulation.
Why has AI safety become such a big issue?
The latest generation of AI systems is moving beyond simple chatbots. AI agents can now interact with software, access information and complete multi-step tasks with limited human intervention.
OpenAI, for example, recently introduced its "dots" autonomous AI agents, which are designed to work across applications and pursue user-defined goals. The development highlights how quickly AI is moving toward systems that can actually perform work rather than simply answer questions.
But greater independence also creates greater risks. An AI agent with access to company systems could potentially make unwanted changes, access sensitive information or behave in unexpected ways.
Companies will have to monitor their own technology
Under the new approach, AI companies will need to build systems capable of identifying potential problems and responding to them. Independent audits are also expected to provide an additional layer of oversight.
The idea is to ensure that AI development does not come at the cost of security and public trust. At the same time, the US administration wants to avoid regulations that could slow down the country's AI industry.
The debate over self-regulation
The biggest question is whether companies should be trusted to police themselves.
Supporters argue that technology companies understand their systems better and can respond to rapidly changing AI risks more quickly than governments. Critics, however, may question whether voluntary commitments provide enough accountability when companies have strong commercial incentives to release increasingly powerful AI products.
The issue is becoming even more important as governments and businesses around the world try to determine how AI should be controlled without preventing innovation.
What does this mean for ordinary users?
For users, stronger internal safety measures could mean more secure AI products, better protection of personal information and greater control over what AI agents are allowed to do.
However, the effectiveness of the new approach will ultimately depend on how seriously companies implement these commitments and how transparent their safety systems become.
AI is no longer just a technology that answers questions. It is increasingly becoming a system that can take actions on behalf of users. As that transition continues, the question of who controls AI—and who is responsible when something goes wrong—will become increasingly important.
