Logan Kirkpatrick, an engineer at Google DeepMind, has countered claims by AI researchers that Google is losing its edge in frontier models (cutting-edge general-purpose AI models).
Regarding "Gemini 3.8 Flash Cyber," a cybersecurity-focused AI model announced by Google on 2026-09-03, Kirkpatrick stated that it possesses superior capabilities compared to models such as Anthropic's "Claude Mythos 5" across many cyber-defense benchmarks and real-world use cases. Gemini 3.8 Flash Cyber is reported to exhibit high performance in vulnerability detection and automated patching.
Additionally, looking toward the future, Kirkpatrick expressed the view that the company's "Gemini 4" model will push the frontiers of the cyber domain.
Source:
- “Google弱体化説”にDeepMindエンジニアが反論 「最新モデルはセキュリティでMythos超え」、「Gemini 4」にも言及 (ITmedia AI+, 2026-09-10)