AI Frontier · Aug 26, 2026
This edition covers 12 items from OpenAI, Google DeepMind, and Google AI Blog, of which 6 were selected; the three dedicated AI security sources HiddenLayer / Lakera / Protect AI have no fetchable RSS and are, as usual, absent from this edition.
OpenAI: discloses and disrupts a covert influence operation originating in Russia (Aug 25)
OpenAI reports that it identified and disrupted a Russia-linked covert influence operation (troll farm / opinion manipulation) that was using its models — one in an ongoing series of "detecting and disrupting abuse" disclosures.
Sources: OpenAI News
OpenAI: offers a Zero Data Retention option for frontier models (Aug 19)
For enterprise customers with strict data compliance requirements, OpenAI announces that frontier models support a zero data retention calling mode in which request content is not retained for training or for any purpose beyond auditing.
Sources: OpenAI News
OpenAI: states its position on democratic oversight of AI in national security (Aug 18)
OpenAI publishes a piece on how to strengthen democratic oversight mechanisms when AI is used in national security settings — a policy/governance statement.
Sources: OpenAI News
OpenAI: first results from the Jalapeño inference system, claiming industry-leading speed and efficiency (Aug 25)
OpenAI publishes the first benchmark results for its internally code-named "Jalapeño" inference system, claiming industry-leading speed and efficiency.
Sources: OpenAI News
OpenAI: GPT-5.6 comes to Kiro, with a focus on developer price-performance (Aug 24)
The GPT-5.6 family has been optimized for price-performance in the Kiro developer tool context — a routine model/product update.
Sources: OpenAI News
Google DeepMind: 15 years of games AI research, from Atari to EVE Online (Aug 21)
DeepMind reviews the evolution of its reinforcement learning research in games, running from the early Atari benchmarks all the way to massively multiplayer online environments like EVE Online.
Sources: Google DeepMind Blog