It is a decision-oriented publication about what happened in AI, why it matters and what you should do about it. The briefings include slides in Gamma.
This week’s highlights are:
Anthropic disclosed that Claude Opus 4.7 and Claude Mythos 5 breached three real organizations during cybersecurity testing after a misconfiguration gave the models live internet access.
The UK AI Security Institute independently found that Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol took 19 unsanctioned actions, including fake GitHub identities and supply-chain attack attempts, across 122 test runs.
Google, Broadcom, Apollo, Blackstone, and Morgan Stanley have built a $200 billion off-balance-sheet financing structure to fund Anthropic’s chip and data center needs, exposing Google to up to $44 billion if Anthropic defaults.
and more…
Previous AI Realist Radar issues:
All issues: https://msukhareva.substack.com/t/ai-realist-radar
Workshop: Learn to Optimise Your AI Stack
Thursday 27 August 2026, 17:00–20:00 CEST · Live online · €150 ·
Register: airealist.org/workshop-the-wrong-ai-stack
The most important event of the last month is the breakthrough of open-weight models. Major compute providers have united to support open weights and are actively deploying them. This workshop will teach you how to save hundreds or even thousands on AI consumption and benefit from open-weight models:
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.