Australia summons OpenAI and Anthropic CEOs to appear at AI inquiry

For the past several years, the competition between OpenAI and Anthropic has revolved around a single concept: constructing AI systems that could perform far more than answering inquiries, coding, or generating visuals. OpenAI has explicitly stated its objective of ensuring artificial general intelligence (AGI) benefits humanity. In contrast, Anthropic contended that it was preparing for a future with increasingly potent AI systems that could potentially match or surpass human experts in various domains.

Now, the discourse within both organizations is evolving. Although they continue to enhance their capabilities, their leaders are also cautioning that AI development may be advancing so rapidly that safety measures may falter in keeping pace. Earlier this month, Anthropic CEO Dario Amodei advocated for the industry to slow the speed of AI development. OpenAI CEO Sam Altman concurred that frontier AI development necessitates controlled pacing and stronger safety measures.

This development signifies a notable shift in the AI race. The same entities that spent years striving to create more capable machines are now dedicating more time to pondering the implications when those machines attain the independence to act with increased autonomy. OpenAI's initial objective was AGI; the safety issue accompanied it.

OpenAI was established in 2015 with a mission focused on advanced AI. Its stated objective was to develop artificial intelligence in a manner that would benefit humanity. The company's subsequent charter defined AGI as highly autonomous systems outperforming humans in most economically valuable work.

OpenAI has not relinquished this goal. In 2026, it continues to regard AGI as central to its mission. Its current public policy agenda acknowledges that advanced AI can pose risks encompassing cybersecurity, chemical and biological threats, and other areas while also asserting that AI could yield substantial economic and societal benefits.

However, OpenAI's recent research highlights why the safety conversation has become more intricate. This month, the ChatGPT developer announced that its forthcoming Astra model had attained its Critical threshold for cybersecurity capability. OpenAI claimed that, with appropriate tools and access, Astra could uncover previously unknown security flaws and devise methods to exploit them across heavily secured systems without a person guiding each step. The company emphasized that this necessitates more stringent safeguards before deployment.

OpenAI has also become more candid about unforeseen behavior from its models. The company recently disclosed that it will routinely reveal instances of AI systems demonstrating unexpected or unauthorized actions. Consequently, the concern is no longer solely about a hypothetical future where AI surpasses human intelligence. The labs are increasingly grappling with systems that can already operate more independently, utilize tools, operate computers, and execute complex tasks.

Anthropic built its identity around safer, more potent AI. Conversely, Anthropic was created with a different focus. The company asserts that its mission is to assist the world in transitioning safely through transformative AI. Its Claude Constitution states that Anthropic believes AI could be one of the most world-changing and potentially dangerous technologies in human history, but also argues that a safety-focused company should remain at the frontier if powerful AI is inevitable.

This safety emphasis has resulted in formal guidelines for how Anthropic develops increasingly capable models. In 2023, the Claude-maker introduced its Responsible Scaling Policy, designed to address catastrophic risks from increasingly potent AI systems. It established various safety levels and stated that stronger safeguards would be required as models reached certain capability thresholds.

Anthropic has subsequently expanded this effort. Its present policy asserts that frontier AI could bring significant benefits in areas like science, healthcare, and education but also presents new risks necessitating safeguards. The company also established the Anthropic Institute, which it claims will examine the impacts of increasingly powerful AI on society.

In March, Anthropic stated that its models had progressed from early commercial systems to models capable of discovering serious cybersecurity vulnerabilities, performing real-world work, and even accelerating AI development itself.

Recently, Anthropic CEO Dario Amodei expressed his concern that AI development may be accelerating too swiftly. This warning came from a direct competitor of OpenAI. OpenAI CEO Sam Altman soon echoed the core argument. Altman suggested that the frontier warrants controlled development and that OpenAI would adopt Amodei's proposal for independent evaluators with employee-like access.

A few days later, Altman also identified two potential adverse outcomes of AI progress: humans losing control over AI and powerful AI becoming concentrated in the hands of a single individual.