Tech chiefs say safety is an engineering issue and reject mandatory laws, requesting that companies be delegated to manage AI risks

The annual conference organized by Salesforce in San Francisco featured an explicit confrontation over the future of AI governance, as Sam Altman, CEO of OpenAI, called for granting tech companies the trust to manage risks themselves amid an absence of imminent regulatory legislation in the United States. Altman told the hundreds of business leaders present that the world has every right to feel afraid of the pace of current tool advancement, noting that imagining pathways for technology to diverge no longer requires broad imagination as it once did, but he also emphasized that his company and the tech sector can lead safety responsibly and keep alignment and safety standards ahead of operational capabilities, while being ready to slow down or stop entirely if needed.

Trust in the Face of Existential WarningsIt formed the focal point of OpenAI chief's first public appearance since a researcher’s resignation post from Anthropic, in which he claimed models could wipe out humanity before the end of the current decade if left unchecked. Although the claim received endorsement from prominent figures such as Elon Musk and Demis Hassabis, head of DeepMind at Google, and prompted Dario Amodei, CEO of Anthropic, to call for slowing development and urging governments to legislate, other leaders deemed the fears exaggerated, and Altman regarded AI labs as ready to self-regulate internally, advising institutions to use the tools in their operations to protect their systems from possible cyber-attacks managed by the same technology.

Jensen Huang, CEO of Nvidia, the world's most valuable semiconductor company, joined the camp opposing new legislation, asserting that tech firms must retain the decision to launch new releases without external interference. Huang said safety is an engineering issue, not a false trade-off between speed of innovation and product protection, noting that the obvious rule is to run at maximum speed and to pause when any organization feels it lacks full control over its product before releasing it to users.

Conversely, the idea of self-regulation faced sharp criticism within industry corridors; Jack Clark, co-founder of Anthropic, said leaving the sector without government oversight is akin to gambling with huge risks. Patrick Hillmann, COO of Logical Intelligence chaired by Jan Luukan, noted a decline in Silicon Valley's public trust compared with formal institutions in Washington, urging companies to demonstrate what they are prepared to stop building rather than merely requesting trust. At the same time, Chris Lihane of OpenAI and Dario Amodei of Anthropic announced voluntary sectoral talks including Google DeepMind to establish shared standards for advanced model safety, whether backed by government or not.

This split in Silicon Valley puts tech executives and institutional leaders in the Gulf, Egypt and the Arab Levant under direct responsibility to reassess their reliance on imported models. With the federal legislative track stalled in the United States and major providers relying on their own judgments to suspend or launch services, banks, government agencies and regional software firms must not rely solely on vendor commitments regarding safety and reliability. The practical impact of this shift calls for establishing independent internal testing frameworks to evaluate model outputs, and building advanced cyber defenses against automated threats as Altman suggested, while diversifying cloud-linking options to avoid operational risks from any sudden freeze decisions made unilaterally by Western labs.

Delegating Safety to Voluntary Engineering DecisionsIt reveals the desire of developer companies to continue harvesting rapid-infrastructure returns away from legal constraints. From a monitoring perspective, betting on goodwill does not create a safe business foundation, and institutions in the region are now concerned with drafting binding technical requirements ahead of integrating any model into their critical operational services.