AI news, 22 August: Claude designs working protein binders at double usual hit rates

Anthropic’s Claude designs lab-validated protein binders. Anthropic reported that its Claude models (Opus 4.8 and Mythos Preview) autonomously designed novel protein binders against 15 biological targets, succeeding on 14 of them. Independent labs Adaptyv Bio and Twist Bioscience synthesized and tested the designs without changes, confirming hundreds of functional binders at hit rates of roughly 23–35 percent—about double the typical 10–15 percent industry average for this early drug-discovery step. Some designs matched or beat prior best results; the work was largely hands-off after an initial prompt. Anthropic has released the dataset publicly. The company frames it as evidence AI can speed parts of biological research, though results varied by target and life-science uses of its strongest models remain restricted. Significance: it shows AI moving from suggestion to physically verified lab work that could accelerate early medicine design.

Nvidia invests in data-center developer Cloverleaf. Nvidia made a minority investment in Cloverleaf Infrastructure to help develop powered sites for AI data centers across the U.S., the companies said Friday. Terms were not disclosed; the Wall Street Journal earlier reported it could run to several hundred million dollars. Cloverleaf will use Nvidia tools for site, power and cooling decisions. The move follows Nvidia’s recent $1.5 billion stake in another developer and underscores how chipmakers are securing the physical infrastructure and electricity needed for AI expansion.

OpenAI cuts frontier model prices for developers. OpenAI announced it is lowering API and credit pricing for its top GPT-5.6 Sol model by more than 20 percent for the next three months, according to Reuters. Standard short-context rates drop to $4 per million input tokens and $20 per million output tokens. Subscription prices stay the same. The cut comes amid competition from Anthropic and Chinese models and follows earlier reductions on smaller GPT-5.6 variants. It aims to make high-end AI more affordable for developers and businesses building applications.

Update: Broadcom AI financing talks grow larger. Reports indicate Broadcom is in talks that could reach $70–100 billion in debt financing for AI chips and capacity, supporting customers including Anthropic, up from the earlier $60 billion-plus figure. Lenders such as Blackstone and Apollo are said to be involved. The scale highlights the enormous capital required for AI infrastructure.

Critical flaw in popular AI framework Ray under active attack. U.S. agency CISA added a severe remote-code-execution vulnerability in the open-source Ray framework (used widely for scaling AI workloads at firms including Amazon, Apple and OpenAI) to its known-exploited list, giving federal agencies just three days to patch. Attackers can target developer laptops via a simple browser visit. A botnet reportedly probed it before public disclosure. Organizations are urged to upgrade immediately.

Orbital data-center startup Starcloud raises big round. Starcloud raised $250 million (valuation $2.3 billion) to build satellite-based AI data centers using Nvidia chips, bringing total funding higher. The company aims to ease terrestrial power and land constraints for compute-heavy AI.