⚡
Brief IA
›

Google unveils Gemini 4 Argon and restricts access to cybersecurity defenders

🛠️ AI Tools·Tom Levy·

Google unveils Gemini 4 Argon and restricts access to cybersecurity defenders

Google unveils Gemini 4 Argon and restricts access to cybersecurity defenders
⚡
Key Takeaways
1Google launches Gemini 4 Argon, initially accessible to "trusted cyber defenders"
2The company announces enhanced safeguards before a broader rollout
3The model is already in use internally and surpasses OpenAI and Anthropic according to Google
4OpenAI has introduced Dots and GPT-6.1 Sol, and is forgoing GPT-6.1 Astra for safety reasons
💡Why it matters — Google is restricting access to its new AI model to a select group, citing its power and the need for increased protections, amid competition and security concerns.
⚡Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

Google introduces Gemini 4 Argon and currently limits its use to "trusted cyber defenders." The company announces additional safeguards before any broader access is granted. Internally, the model is already in use and outperforms competitors on benchmarks shared by Google. The launch comes a day after OpenAI's DevDay, marked by Dots, GPT-6.1 Sol, and the abandonment of GPT-6.1 Astra for security reasons.

Google Delays Broad Access and Plans Safeguards

Google plans to enhance protective measures before a wider rollout of Gemini 4 Argon, citing the need to defend against abuse, prompt injection attacks, and monitoring for misalignments. The launch occurs a day after OpenAI's DevDay, during which OpenAI introduced the Dots agent and the GPT-6.1 Sol model, while announcing this week that it would not release GPT-6.1 Astra due to security concerns. Koray Kavukcuoglu, appointed head of DeepMind in August, presents Gemini 4 Argon as a model whose access remains restricted to "trusted cyber defenders," with a gradual expansion announced.

The Model Claims High Performance and Is Already in Internal Use

According to Koray Kavukcuoglu, Gemini 4 Argon targets complex workflows in software engineering, enterprise knowledge, particularly in legal and finance sectors, as well as in cybersecurity. He notes that the model is already powering internal processes at Google and has contributed to large-scale code migrations. Google has released a table of numerous benchmarks and claims that Gemini 4 outperforms models from OpenAI and Anthropic. Koray Kavukcuoglu, Chief AI Architect and Senior Vice President of Google DeepMind, emphasizes these positioning elements.

Restricted Access and Ongoing Government Evaluation Process

Initial access to Gemini 4 Argon is limited to a set of trusted cyber defenders. Koray Kavukcuoglu specifies that Google is participating in the U.S. government's voluntary process for early access to models, alongside a gradual scaling up. Earlier this week, discussions on X mentioned alleged leaked benchmarks related to the model. Google officially presented Gemini 4 Argon today.

⚡

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.