AI Brief: 2026-08-24

Executive Summary

Anthropic detailed its SynthID-derived watermarking architecture for Claude text outputs, ensuring direct compliance with incoming EU AI Act transparency mandates.
Anthropic expanded Claude Mythos 5 access to cybersecurity teams, enabling automated enterprise codebase vulnerability scans, threat detection, and advanced malware analysis.
OpenAI slashed GPT-5.6 Sol API rates by over 20% to $4 input and $20 output per million tokens under a promotional three-month tier.
OpenAI deployed unpersonalised ChatGPT Ads across 31 European territories, targeting Free and Go subscription tiers amidst rising monetisation pressures.
Meta launched its Llama 4 multimodal suite (Scout and Maverick) with a 10-million-token context window, alongside Muse Glimmer—a 30B open-weight model engineered for local agent execution.

Details revealed on Claude text watermarking using SynthID-family technology for EU compliance.
Claude Mythos 5 released to security teams for vulnerability scanning and malware analysis.

GPT-5.6 Sol API prices reduced to $4/$20 per million tokens for three months.
ChatGPT Ads rolled out across 31 European markets without personalization at launch.
OpenAI called for strengthening California's SB 53 AI safety framework.

Llama 4 released with Scout and Maverick models featuring a 10M token context window.
Muse Glimmer launched as a 30B parameter open-weight model for local multimodal AI agents.

Gemini 3.7 integrated into search AI mode; updated Gemini usage limits and pay-as-you-go options rolled out.

Introduced Agentic Search for higher retrieval accuracy and Shieldstral 3B open-weight moderation model.

GPT-5.6 Sol — OpenAI. OpenAI reduced promotional API pricing for its flagship GPT-5.6 Sol model by over 20% to $4 input and $20 output per million tokens through November 21. Read more
Claude Mythos 5 — Anthropic. Anthropic enabled Claude Mythos 5 access for enterprise security teams to perform automated codebase vulnerability scanning and threat detection. Read more
Llama 4 — Meta. Meta introduced the Llama 4 family, including Scout and Maverick models, supporting multimodal workloads with up to a 10 million token context window. Read more
Muse Glimmer — Meta. Meta released a 30B open-weight Apache 2.0 multimodal model designed specifically for running autonomous AI agents on local infrastructure. Read more

Interpretation, not reported fact. Each point follows from a source listed below — follow the links and judge for yourself.

Widespread watermarking adoption may drive demand for automated compliance audit tooling. (Basis: Anthropic's implementation of SynthID-family watermarking to meet EU AI Act requirements.)
Price competition among frontier model providers will likely accelerate multi-model routing strategies. (Basis: OpenAI's 20%+ price reduction on GPT-5.6 Sol.)