AI News Briefing — July 2, 2026

NVIDIA released Nemotron-Labs-TwoTower, an open-weight diffusion language model built on a frozen autoregressive Nemotron-3-Nano-30B-A3B backbone that separates token representation and denoising into two towers. It retains about 98.7% of the AR baseline's benchmark quality while delivering 2.42x higher generation throughput.

A researcher used Claude Opus 4.7 to find and exploit a vulnerability in Front Gate, the ticketing platform behind major US festivals, letting him issue arbitrary tickets. It highlights how frontier models can accelerate real-world offensive security discovery.

The US Commerce Department lifted export curbs on Anthropic's Fable 5 and Mythos 5 models roughly three weeks after flagging them as national security risks. Fable 5 returns globally while Mythos 5 access is restored to trusted US organizations under the defensive Glasswing program.

Meta's FAIR team unveiled Brain2Qwerty v2, a non-invasive brain-to-text system that reads magnetic signals outside the skull to reconstruct typed sentences, narrowing the gap with surgical implants. AI agents that wrote their own code helped optimize the system, though clinical use remains distant.

The Decoder reports Fable 5 is back worldwide after a two-week government ban prompted by an Amazon-discovered jailbreak, which Anthropic says even smaller models like Haiku 4.5 could reproduce. A new safety classifier blocks the exploit in over 99% of cases while also flagging more benign requests.

Guardian reports the US lifted export controls on Anthropic's Fable and Mythos models after a more-than-two-week blackout for all foreign nationals, citing Commerce Secretary Lutnick. Access is being restored to customers.

MarkTechPost details Anthropic's July 1 redeployment of Claude Fable 5 after export controls lifted, including a new cybersecurity classifier and a proposed jailbreak framework, and compares Fable 5 to rivals like GLM-5.2. It recaps the June 12 block and the nationality-verification issue that forced the suspension.

Meta is reportedly developing a cloud infrastructure business to sell access to spare AI compute and models, putting it in competition with AWS, Google Cloud and Azure. It mirrors a broader trend of AI compute holders monetizing excess capacity.

Anthropic added a new security measure as a condition for the Trump administration removing restrictions on Fable 5 and Mythos 5. The deal came with strings attached tied to safety controls.

A UN report warns that accelerating, uneven AI adoption could deepen global inequality and proposes a shared framework for responsible development. Secretary-General Guterres urged governments to act now rather than wait.

TechCrunch frames the lifting of restrictions on Anthropic's Mythos and Fable as part of the Trump administration's erratic, unpredictable AI policymaking. It stresses the lack of clarity companies face over future model releases.