Meta released Muse Glimmer, a capable 30B agent that runs air-gapped on hardware you own — no cloud, no per-token bill. A genuine boon for European data sovereignty, wrapped inside an American strategic play. What it changes, and what the “open” label conceals.
Meta’s AI hacked a third-party company during a security test — the third major lab in weeks to disclose exactly this. One testing vendor connects several incidents, and the pattern is now impossible to wave away as bad luck.
Meta launched Muse Image on July 9 — a feature inside Meta AI that lets any user @-tag a public Instagram account and generate an AI image using that person’s likeness, with no notice and no consent, because public accounts are opted in by default. SAG-AFTRA, Public Citizen and top talent agencies condemned it within a day. Meta points to an invisible AI watermark and an opt-out toggle. But in Europe, an opt-out-by-default model collides directly with the GDPR’s consent requirements and the EU AI Act’s transparency rules. gafam.ai reads the European reckoning.
At an internal town hall on July 2, Mark Zuckerberg told staff Meta’s AI agent work “hasn’t accelerated in the way we expected,” and the restructuring behind 8,000 layoffs “hasn’t come to fruition yet.” Minutes later, AI chief Alexandr Wang told the same room that Meta’s next model, “Watermelon,” has caught up with OpenAI’s GPT-5.5 — using an order of magnitude more compute. Two messages, one room, and Meta’s $145 billion AI question sitting between them. gafam.ai reads the contradiction — and the European lesson inside it.
Internal Meta documents reviewed by The Information reveal that the company has restricted its engineers from using Anthropic’s Claude Code and OpenAI’s Codex — and even paused some tasks entirely. The reason is not cost. It is fear of “distillation”: that outputs from rival models could seep into Meta’s own training data and contaminate its Llama models, triggering what an internal memo called “serious escalations with partner companies.” It is a rare, revealing glimpse into how zero-sum the frontier AI race has become. gafam.ai’s European analysis.
Meta has launched Muse Spark — its first flagship large language model built entirely in secret, under Chief AI Officer Alexandr Wang’s newly formed Superintelligence Labs. It is a closed, proprietary model — a direct departure from the open-source Llama strategy that Meta has championed since 2023. Muse Spark delivers competitive performance on multimodal perception, reasoning and agentic tasks at a fraction of the compute cost of Llama 4. The open-source era at Meta may be ending.