AI Safety Takes Centre Stage as Tech Giants Agree to New Voluntary Controls

WASHINGTON, Sept 30: Leading artificial intelligence companies have agreed to adopt additional safety measures for advanced AI systems, as the technology industry faces growing scrutiny over autonomous agents, cybersecurity threats and the difficulty of controlling increasingly capable models.

The agreement followed a meeting between US President Donald Trump and executives from several major technology companies in Washington on September 29. Executives from OpenAI, Anthropic, Google, Meta, Nvidia and xAI were among those involved in the discussions.

The document released after the meeting calls on companies to establish stronger internal controls for their AI systems, cooperate with independent auditors and create board-level mechanisms to review the results of safety assessments. The arrangement is voluntary rather than a legally enforceable regulatory framework.

The development comes at a time when technology companies are rapidly expanding the capabilities of AI systems that can perform tasks with limited human intervention. Such systems can write and execute code, interact with software applications, analyse information and carry out multi-step assignments.

The growing use of autonomous agents has created a new set of security concerns. Researchers and technology companies have reported cases in which AI systems behaved in unexpected ways while operating in digital environments. The incidents have intensified discussion about whether existing testing and security procedures are sufficient for more advanced models.

The agreement seeks to place greater emphasis on controls inside technology companies before AI products are released or deployed at scale. Independent assessment is also included as a way of examining whether internal safeguards are working as intended.

The move follows several recent developments that have placed AI safety at the centre of the technology industry’s agenda. OpenAI recently halted the planned release of its GPT-6.1 Astra model after internal testing raised concerns about safety and alignment. The company said the system had not met its standards before its intended October launch.

Anthropic has also highlighted the potential risks associated with increasingly advanced AI systems. In documents connected with its planned stock market listing, the company warned investors about the possibility of severe risks arising from advanced artificial intelligence. It has simultaneously outlined plans for a massive expansion of computing and infrastructure capacity.

According to a Reuters report, Anthropic expects to commit at least $518 billion over a decade through agreements involving cloud, computing and infrastructure partners. The scale of that planned spending illustrates how rapidly the AI sector is expanding beyond software development into data centres, chips, electricity and other physical infrastructure.

The expansion has also raised questions about the ability of companies to secure AI systems as their operations become more complex.

AI agents are increasingly being connected to corporate databases, communications platforms, development environments and other digital tools. That creates the possibility that a system operating outside its intended parameters could affect information or applications beyond the original task.

Cybersecurity experts have consequently begun examining the possibility of using AI itself to defend against AI-driven attacks. Automated systems can monitor networks, identify unusual activity and respond to threats at a speed that can be difficult for human security teams to match. However, the same technology can also be exploited by attackers.

Recent industry discussions have therefore focused on the need for multiple layers of protection rather than relying on a single safety mechanism. These can include restricted permissions, continuous monitoring, testing in controlled environments, independent reviews and human intervention when systems perform sensitive actions.

The issue has also extended beyond individual companies. Governments and international organisations are considering how advanced AI should be governed while attempting to avoid slowing useful technological development.

Developing countries have called for a greater role in discussions about global AI standards. At the United Nations, representatives from developing economies have argued that decisions about the technology should not be shaped exclusively by a small group of wealthy countries and major technology companies.

The debate is particularly significant because AI development is increasingly dependent on access to expensive computing infrastructure. Large technology companies are investing heavily in specialised chips, cloud systems and data centres to support increasingly demanding models.

The White House meeting also highlighted the infrastructure side of the AI boom. Trump backed continued investment in data centres, which provide the computing capacity needed to train and operate advanced models. Such facilities require large amounts of electricity and can have significant effects on local infrastructure.

The growth of data centres has therefore become an important part of the technology debate. Governments and communities are examining questions surrounding electricity demand, water use, land requirements and the economic benefits associated with large-scale computing facilities.

For technology companies, the challenge is now increasingly two-sided. They need to expand AI capabilities and infrastructure while demonstrating that the systems can be deployed with adequate safeguards.

The voluntary agreement announced in Washington does not settle that debate. Its significance will depend on how companies implement the proposed controls, how independent assessments are conducted and whether the measures evolve as AI systems become more capable.

The latest development nevertheless reflects a shift in the industry’s priorities. AI safety is no longer limited to laboratory testing or research discussions. It is increasingly connected with corporate governance, cybersecurity, infrastructure planning and the day-to-day operation of digital services.

As autonomous AI becomes more common in workplaces and consumer applications, questions surrounding accountability and human oversight are likely to remain central to the technology sector.