The Yardstick Feed
AI updates, frontier-model releases, and standards work.
A weekly digest for B2B AI buyers: capability and safety announcements from frontier labs, AI standards and regulatory updates, agent-reliability research, and Yardstick methodology notes. Sourced from a fixed list of labs, regulators, and research institutions.
-
Testing Jev in High Volume Retrieval and Routing
We tested Jev, TypeSafe's typed-choice classifier, as a router that decides which vendor-dossier claims can skip a second review. 590 live claims, 127 accepted with none judged bad by a blind reviewer; 372 historical claims; planted false claims; prompt injection. Scripts and anonymized data are published with the article.
By Yardstick Research
-
We audited our own AI for hallucinations: the method, the data, and the script that checks our math
25 models, the 30 hardest companies we cover, 454 audited runs, every claim checked by deterministic code against frozen copies of real source pages. No model produced zero hallucinations, and the gate caught every one the winning pipeline produced before anything published. The whitepaper, the full anonymized dataset, and a verification script that recomputes every number are free to share and republish under CC BY 4.0.
By Yardstick Research
-
AI Updates, Week of May 22, 2026
Frontier-lab consolidation and oversight signal both showed up this week. Cohere shipped Command A+ and announced two strategic MOUs the same day, DeepMind released its Co-Scientist research-agent system, AWS Bedrock unlocked per-request cost attribution, and the UK AI Security Institute published its Frontier AI Trends Report.
Frontier labs
-
Introducing Command A+ Cohere Blog
An open-weights mixture-of-experts model built for high-performance agentic tasks that can be privately deployed on as little as two H100 GPUs. The pitch: enterprise-grade reasoning and multimodal capability without the inference-cost profile of a frontier dense model.
Read at cohere.com → -
Co-Scientist: A multi-agent AI partner to accelerate research Google DeepMind
A multi-agent system built on Gemini that iteratively generates, debates, and refines novel hypotheses for life-sciences research. Positioned as a collaborator for working scientists rather than a one-shot reasoning engine.
Read at deepmind.google → -
Cohere acquires Reliant AI to expand sovereign enterprise AI Cohere Blog
Reliant AI is a Montreal- and Berlin-based biopharma AI company. The acquisition extends Cohere's sovereign-AI positioning into healthcare-vertical workflows where data residency and compliance shape vendor selection more than raw model capability.
Read at cohere.com →
Infrastructure & platforms
-
Amazon Bedrock expands support for request-level usage attribution AWS What's New
Customers can now tag InvokeModel and InvokeModelWithResponseStream calls with team, application, environment, and experiment metadata for per-request cost reporting. Closes a long-standing gap for enterprises that need to attribute Bedrock spend across business units.
Read at aws.amazon.com →
Standards & regulation
-
Frontier AI Trends Report UK AI Security Institute
An assessment of the AI oversight landscape, its robustness to capability advances, and the pathways that could lead to its degradation. Pairs with NIST's RMF as buyer-facing evidence on whether a vendor's claims about model safety hold up to independent evaluation.
Read at aisi.gov.uk →
By Yardstick Research
-
-
AI Updates, Week of May 15, 2026
Distribution and oversight dominated the week. OpenAI stood up a Deployment Company to embed engineers with enterprise customers, the UK AI Security Institute shipped two safety papers, and Anthropic announced a four-year, $200M partnership with the Gates Foundation.
Frontier labs
-
OpenAI launches the OpenAI Deployment Company OpenAI
A new arm of OpenAI focused on embedding Forward Deployed Engineers with enterprise customers, paired with the acquisition of Tomoro. Signals an industry shift toward services-and-models bundles rather than model-API-only sales.
Read at openai.com → -
Anthropic forms $200 million partnership with the Gates Foundation Anthropic News
Four-year commitment of grant funding, Claude usage credits, and technical support across global health, life sciences, education, and economic mobility. Foundation-grade enterprises now have a defined path to Claude deployment with vendor support included.
Read at anthropic.com →
Standards & regulation
-
Alignment research paper UK AI Security Institute
AISI's alignment work probes whether frontier-model behavior actually matches stated objectives under adversarial conditions. Useful as procurement reference when a vendor claims their agent is safe by design.
Read at aisi.gov.uk → -
Cyber autonomous systems capabilities research UK AI Security Institute
AISI's narrow-cyber suite finds the length of tasks frontier models can autonomously complete in cybersecurity scenarios is doubling every few months. Direct implication for any AI-buyer doing threat modeling, the offensive-capability curve outpaces typical procurement cycles.
Read at aisi.gov.uk →
By Yardstick Research
-
Tip us on a source
A source we should be tracking?
Email a link to a frontier-lab blog, a standards body, or a research feed worth adding to the source list. Vendor pitches belong on the submit-vendor page instead.
hello@yardstickresearch.app
Email a tipTake the free 4-minute readiness audit.
Get your score, peer benchmarks, and three tailored vendor recommendations. No email required to see your results.