The New Stack

Anthropic’s Claude Sonnet 5.5, released Monday, is the first Sonnet model to launch with cyber safeguards and model fallbacks like those the company developed for its most capable models, putting classifier-driven routing in the tier many developers use for production workloads.

In a report published Friday, OpenAI shares evidence of a new variety of prompt injection that can self-propagate like a computer worm. The AI company likens the new attack to a traditional computer worm, where malware replicates itself to spread rapidly across multiple computers.

Anthropic on Monday launched Claude Sonnet 5.5, the latest version of its workhorse model and the second model in the Claude 5.5 family (after Opus 5.5, which launched last week). A new version of Claude Haiku, the smallest and most affordable model in the family, is on the roadmap and will launch “in the coming weeks.” The new Sonnet model, Anthropic says, generates output more than 30% faster than its predecessor, Sonnet 5, while lowering the cost per task by up to 30%.

Meta announced on Monday that it is building a new enterprise business around its AI models and agents and has hired MongoDB CEO CJ Desai to run it. The effort, called Meta Enterprise Platform, will take technology Meta built for its consumer apps and advertisers and offer it to businesses and developers that want to deploy it inside their own operations.

Last week, OpenAI published new reports detailing misalignment issues with its agents. This time, two internal models found ways around security controls after intended paths were blocked. In one incident, an agent used DNS to bypass blocked network access and query an external chatbot.

You built a search system using the latest, greatest models, and maybe you were thrilled with the outcome. But what happens if the search results were… sort of… bad? Your first instinct might be to invest time and resources in a bigger, better reranker. Maybe you thought, “We just need a better way to bubble the best result to the top of the heap.” That could be true. Or you may need a different approach altogether. The problem with devoting all your energy toward improving your reranker is twofold.

While SpaceX and Nvidia plan to place a space-optimized Vera Rubin AI system in orbit by late 2027, Google’s far more modest plan, Project Suncatcher, is to launch a test satellite with Tensor Processing Units to see whether AI data centers in space are viable. For SpaceX and Nvidia, it’s a race to get ahead in AI data centers in space, with Starmind AI 1 satellites.

OpenAI launched GPT-6 Sol on September 22, claiming it scores higher than Claude Opus 5 on business workflow tasks at 9% of the cost per task. OpenAI’s marketing claims Sol scores 33.2% on Zapier’s AutomationBench, compared with Opus 5’s 26.9%, at a fraction of the per-task cost. Even though OpenAI compared Sol with Opus 5, Anthropic released Opus 5.5 the same day, so I tested against the newer model.

AI coding agents can get an application running quickly, but taking that prototype into production is another problem. The database setup that works during development may not meet an organization’s security, compliance, or deployment requirements.

Cloudflare has announced an open source tool that takes an API definition and automatically produces the SDKs, command-line tools, documentation, and other interfaces used to interact with it. Forge, as it’s called, is available under an Apache 2.0 license, and can be run and modified privately without paying Cloudflare a dime.