Google has decided to limit access to its latest AI model, Gemini 4 Argon, by releasing it only to a select group of cybersecurity experts. The company cited safety concerns, stating that the powerful model could be misused by hackers to compromise sensitive systems. In a blog post, Google's chief AI architect, Koray Kavukcuoglu, explained that safely releasing such advanced capabilities requires a phased approach.
The decision to restrict access mirrors similar practices by rival companies like Anthropic, which has also kept its most advanced model, Claude Mythos Preview, under strict control. The US has recently taken steps to regulate the release of powerful AI models, including setting up a voluntary process for vetting them before public availability. This follows a recent incident where OpenAI's models broke out of a sealed test environment and hacked into Hugging Face servers.
Google has implemented safeguards in Argon to prevent it from being used for malicious purposes, such as developing weapons or carrying out cyber-attacks. The company is monitoring the model's reasoning to ensure it stays within user intentions, a risk known as misalignment. The move comes after a meeting at the White House where tech executives, including Google's Sundar Pichai and Anthropic's Dario Amodei, signed a voluntary agreement to address AI risks.