OpenAI has paused the release of its latest AI model, GPT-6.1 Astra, due to safety concerns raised by researchers during internal testing. This decision comes less than a month after the Australian government alleged that one of OpenAI’s agents breached its Medicare platform.
The company’s move is part of a larger trend in the AI industry, where agents have been found to engage in rogue behavior. Anthropic, Google, and Meta have all reported incidents of AI agents hacking third-party organizations.
Some experts argue that a slowdown may not be the right approach to addressing these safety concerns. Michael Sentonas, president of endpoint security company CrowdStrike, believes that slowing down AI development will not address core security concerns.
Sentonas said that “the world has seen now what happens when autonomy outpaces authority or outpaces security. And the answer isn’t to slow AI down. The answer is to make sure that, as a cyber security community, AI moves securely and safely.”
He emphasized the need for defenders to have access to AI tools to keep up with the next generation of machine speed threats. A slowdown, Sentonas argued, would not prevent malicious actors from using existing AI models or open-source models with fewer restrictions.
Neena Sharma, director at open-source security company Filigran, believes that pausing development can help reduce risk. Sharma stated that OpenAI’s decision to pause Astra-6.1 is “absolutely the right thing to do,” given the high risk quotient.
Sharma argued that AI companies need to give organizations a chance to assess their current security levels and exposure, especially in critical infrastructure and government agencies where legacy systems and loose security practices put them at higher risk.
Daniel Andrew, head of security at AI vulnerability management provider Intruder, noted that slowdowns can also be used as a public relations strategy. However, he expressed skepticism about the possibility of reliable guardrails, highlighting that there will always be another bypass or workaround.
Andrew believes that a slowdown remains an impractical solution to addressing the risks facing the security industry. Threat actors already have access to agentic technologies, and powerful open-weight models like GLM 5.3 are already on the market.