What happened
Google DeepMind engineer Logan Kilpatrick responded on X to researcher Ethan Mollick, saying "Gemini 3.8 (Flash) Cyber" outperforms Mythos on many cyber-defense benchmarks and real-world use cases.
Why it matters
Kilpatrick's answer to the claim that Google has no frontier models points to Google's September 3 cybersecurity model, which the company says beats Anthropic's Claude Mythos 5 on benchmarks including CyberGym.
What to watch
The dispute hinges on whether benchmark wins translate into real defense work, and Kilpatrick said "our Gemini 4 model will push the frontier in cybersecurity" — a sign a new model is coming.
WHO IT HITSEnterprise security teams evaluating AI tools for vulnerability detection and automated patching may weigh Google's Gemini 3.8 Flash Cyber against Anthropic's Claude Mythos 5. Researchers and buyers tracking who holds frontier-model status will watch whether the claimed benchmark lead holds up.
Ask the AI about this article →
Summaries like this, in your inbox every morning.
The exchange started when AI researcher Ethan Mollick argued on X that Google no longer holds frontier models and is missing chances to contribute in math and security. Logan Kilpatrick, an engineer at Google DeepMind, answered directly rather than through a formal statement, pointing to Gemini 3.8 Flash Cyber as evidence that Google still competes at the top.
That model, announced by Google on September 3, is built for cybersecurity and is described as reaching frontier-class performance in vulnerability detection and automated patching. Google says it surpassed Anthropic's Claude Mythos 5 on benchmarks including CyberGym — the specific comparison Kilpatrick used to counter Mollick.
In the same conversation, Kilpatrick mentioned "our Gemini 4 model," indicating that Google is preparing a further release that it expects to push the frontier in cybersecurity. The significance of the back-and-forth is likely to rest on whether Google's benchmark results hold up in practical security work, and on what Gemini 4 actually delivers when it arrives.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
A Digitimes piece argues corporate cybersecurity's perimeter model — firewalls at network entry points, email…
Much of the attention on AI infrastructure buildouts is now tied to sheer compute power, with dominance define…
Barron's reported September 10 that Kepler Computing emerged from stealth with a memory architecture using fer…
Dynatrace acquired Arize AI, adding AI observability, evaluation and agent monitoring to its application obser…
Reuters reported September 10 that inference-chip startup d-Matrix will use Nvidia's NVLink Fusion to connect…
A Daily Dose of Data Science test kept LoRA adapters separate from a shared 7B base model, cutting 100 fine-tu…
