YFarmX logoYFarmX

Frontier technology news: AI, crypto and quantum

Explore desks

State of playLatest market update

Updated

Latest

Open the Newswire →
Editorial illustration: MODEL SAFEGUARDS over a heavy turnstile gate with one bar painted red, with the Anthropic, OpenAI and Google logos

AINew release

Model safeguards on security requests: what each lab puts in front of the model

What each frontier lab puts between a security request and its model in 2026: Anthropic's typed refusals with a fallback to Opus 4.8 and published block rates, OpenAI's two Daybreak tiers, Google's access programmes, and the licences binding the open-weight models, read from the labs' own documents.

AINew release

Google Big Sleep: the agent finder, 26 attributed entries and the SQLite origin

Google Big Sleep is the vulnerability agent built by Google DeepMind and Project Zero, announced 1 November 2024 with a stack buffer underflow in SQLite. Eighteen CVE records name it in their credits and seven more carry its credit in Apple's and Google's advisories, each checked at the record in September 2026.

AINew release

Jev (TypeSafe): a model that answers typed questions with probabilities

TypeSafe opened early access to Jev on 15 September 2026, the first of what it calls System One models. It takes a block of state plus typed questions and returns a choice, a boolean or a score, each with a calibrated probability, on OpenRouter's decisions route at $0.042 per million input tokens.

Risk trackers

Tools

Tech explainers

Explained

The models, assets and machines behind the news, across all three desks.

YFarmX logo

Frontier Tech
Events

Navigate crypto, AI and quantum technology conferences worldwide.

YFarmX logo

Arcade

Four hand-built games. High scores stay on your device.

Play arcade

Look it up. Every model, coin and qubit.

Every system, on the record

Or look a term up in the Glossary