Google limits access to powerful AI model over safety concerns

Alphabet's Google unveiled a new top-tier artificial intelligence (AI) model called "Argon" on Wednesday, designed to bolster its Gemini 4 generation of models and compete with rivals Anthropic and OpenAI. However, Google decided to limit the model's public release for now, only sharing it with a select group of cybersecurity experts to prevent potential misuse by hackers. Google's chief AI architect, Koray Kavukcuoglu, stated in a blog post that releasing frontier capabilities at this level requires a phased approach.

The new model is larger in size compared to Google's previous line of advanced "Pro" models, according to a company spokesperson. Google claims Argon to be its most performant model yet, built to handle complex workloads, and compares it to frontier models like OpenAI's Astra and Anthropic's Opus on key coding and cyber benchmarks.

Google provided the model to select cybersecurity partners and is participating in the Trump administration's voluntary process for pre-release model access. The timeline for the model's public release has not been disclosed.

While Argon performed better than OpenAI's Astra and Anthropic's Opus in several industry benchmarks, it lagged behind in certain metrics, including two of the four coding-related benchmarks. Google highlighted Argon's proficiency in complex tasks, such as software engineering, legal and financial work, and cyber defense, noting its superior ability to identify and rectify critical software flaws.

Early testers used the model to discover a flaw in hospital software that exposed sensitive personal data, a flaw that other advanced models had missed, Google reported.

The new AI model is designed to refuse requests that could facilitate cyberattacks or help develop chemical, biological, or nuclear weapons. Similar safeguards have been implemented in models from Anthropic and ChatGPT-maker OpenAI. Google's cautious rollout comes a day after U.S. President Donald Trump hosted top tech executives, including Google CEO Sundar Pichai and Anthropic's Dario Amodei, at the White House, where they signed a voluntary agreement to address the risks associated with their AI systems.

Cybersecurity experts express concern that the state-of-the-art technology could be exploited to hack banks, hospitals, and government systems. The cautious approach mirrors the strategy employed by rival Anthropic, which has restricted its most advanced model, Claude Mythos Preview, to a small number of trusted organizations.