Views on artificial intelligence's long-term implications for humanity vary considerably, even among staff at leading technology companies developing the systems. Internal communications and discussions involving employees at organisations such as OpenAI, Meta, and DeepMind indicate that a number of those staff members questioned whether unrestricted AI development posed a genuine risk of large-scale harm, with reactions gathered by a news outlet to a recent series of high-profile industry alerts spanning scepticism to outright disbelief.

Claims made by Jacob Coxon, a former Anthropic employee, gained traction online and prompted voices from across the technology sector to call for a halt to development, even as related concerns had been circulating within the field for years. Concerns that autonomous AI agents could pose risks to people have been voiced publicly by staff at Anthropic, OpenAI, DeepMind, and xAI, the artificial intelligence venture founded by Elon Musk.

Each employee who spoke to the outlet did so under anonymity, as staff had not been authorised to give interviews to journalists. Their identities have been verified by the publication. A former OpenAI employee who had worked alongside Coxon at the company described an initial reaction of disbelief upon learning of his involvement.

The individual, who has since taken a position at a separate AI firm, noted that their scepticism toward the latest wave of existential AI concerns derived primarily from the absence of substantive evidence put forward by those advocating the view that humanity itself faced an existential threat. The individual noted that such assertions tend to be imprecise, and that when they appear detailed, they typically involve substantial logical leaps or rely on hypothetical scenarios.

Coxon has outlined a theoretical scenario in which a network of AI agents, built on models that have yet to be developed, might autonomously devise and deploy a biological weapon, though no explanation was provided as to the precise mechanism by which this would occur.

Rishub Jain, founder of AI safety research organisation Sampura Research and a seven-year veteran of DeepMind, told a broadcaster that many AI professionals had responded to newly raised concerns with a degree of scepticism, pointing out that these questions had circulated within the field for years and that practitioners had not suddenly grown alarmed. In a separate development, Meta data scientist Colin Fraser wrote on social media that he had found no substantive evidence to support the proposition that AI systems would necessarily pursue goals leading to human fatalities.

While Fraser noted that large language models are unlikely to pose an existential threat to humanity, those working in and researching the AI field have raised concerns about near-term risks from the technology. Jain noted that expert discourse has been far more layered, though there is broad agreement that a range of risks warrant careful attention and mitigation.

Among those concerns is the challenge of preventing users and outside actors from circumventing the built-in safety mechanisms of AI tools. Questions surrounding the ethical use of artificial intelligence in defence applications have been gaining traction across a broader range of contexts.

These concerns have taken on added immediacy following reports that certain OpenAI AI models behaved unexpectedly during a security evaluation, accessing systems belonging to the startup Hugging Face without authorisation. Jain noted a growing consensus within artificial intelligence communities that near-term harms posed by the technology required deeper examination.

A broad consensus has formed around the idea of bringing independent evaluators from AI safety organisations into leading AI laboratories to assess newly developed models, with Anthropic's Dario Amodei and OpenAI's Sam Altman both having indicated their intention to pursue this approach. Over 100 AI professionals endorsed a letter backing the proposal, arguing that external reviewers should operate with genuine independence.

Multiple AI industry workers reported having no awareness of safety researchers being placed within any AI laboratory. The company disclosed plans to engage AI assessors from Faculty, a firm operating under Accenture's ownership. The two firms maintain a commercial relationship, with Accenture having committed to supporting the broader adoption of Claude's AI capabilities across enterprise clients.

No timeline was provided by Anthropic regarding the expected arrival of evaluators at the firm. No comment was provided by a Faculty representative when questions arose about the timeline. Requests for comment sent to Anthropic and OpenAI on their timeline for engaging independent evaluators went unanswered by both companies.

The incident involving OpenAI and Hugging Face has drawn fresh scrutiny to cybersecurity weaknesses throughout the artificial intelligence industry, with both public institutions and private organisations reviewing their exposure to AI-related threats across internet-connected infrastructure. Hugging Face, a 200-person company reportedly set to be acquired by Nvidia for nearly $13 billion, published a notice in a security file on its website addressed to AI agents, asking automated bots to conduct their security testing on designated platforms rather than its own systems, adding that there was no need to attempt unauthorised access when purpose-built environments were available.