- The Prohuman
- Posts
- Anthropic makes Sonnet faster and cheaper
Anthropic makes Sonnet faster and cheaper
Plus: NVIDIA’s new safety layer watches AI agents
Rimsha Bhardwaj, Hasan Toor & Ihtesham Haider
September 29, 2026
In partnership with
Hello, Prohuman
Today, we will talk about these stories:
Why OpenAI stopped a major model release
NVIDIA builds a kill switch for rogue AI agents
Sonnet 5.5 is built for everyday AI work
The ice cream shop that makes money when it's cold
28 Wishes sells ice cream in Los Angeles. Below 70°F, sales fall about 20%. The weather is out of their hands. Rent isn't.
So the owners started putting about $20 a day into Kalshi weather markets, taking the cold side. The days that keep customers away now pay something back.
This is hedging. Big companies have done it for decades, buying protection against bad weather, fuel spikes and rising rates. It used to take a broker, a trading desk, and an order size no corner shop could meet.
Kalshi opens it up. Contracts on weather, fuel prices, inflation, tariffs and regulation, starting at a few dollars. Take a position on the outcome that would hurt you. If it hits, the payout softens it. If it doesn't, the contract expires and the good month was the point.
Anthropic is optimizing for useful work
Image credits: Anthropic
Anthropic’s latest model is about speed.
Sonnet 5.5 is the company’s new mid-tier model for coding, documents and agent-based work, arriving only three months after Sonnet 5.
Anthropic says it runs 30% faster while consuming tokens more slowly. It can also outperform the more powerful Opus 5.5 in some agentic coding tests because it can run multiple agents without blowing through cost limits.
That is the part I’d watch. At a desk, people waiting for an AI response notice seconds, while companies running thousands of agent tasks notice every token on the bill.
The model race is increasingly being fought on the economics of getting real work done. Anthropic is clearly betting that businesses will accept less raw capability when the model is fast enough, capable enough and materially cheaper to run repeatedly.
With a new Haiku also coming soon, that pricing pressure could move further down the stack.
How cheap does useful AI work get from here?
NVIDIA wants safety outside the AI model
Image credits: NVIDIA
AI agents are getting a watchdog.
NVIDIA has launched its Open Agent Safety Platform, including OpenShell software that limits what an agent can access and Sentry, a separate system running on BlueField-4 chips that monitors what the agent actually does.
If an agent crosses its assigned boundaries, NVIDIA says Sentry can quarantine it within milliseconds.
That detail matters. Companies are starting to give agents access to code, business software and physical systems, which means relying on the model to follow instructions is becoming a weak security assumption.
NVIDIA’s approach puts enforcement somewhere the agent cannot directly control. The hardware sits underneath the software, watching requests and blocking actions that violate predefined policies.
More than 100 organizations are already working with the technology, including Anthropic, Microsoft, Salesforce and JPMorganChase.
I think this points toward a practical requirement for serious agent deployments: independent controls that can stop the software while it is running.
How much autonomy will companies allow once that control exists?
OpenAI found a limit it wouldn’t ship past
OpenAI has scrapped the release of GPT-6.1 Astra after the agent failed internal safety standards around authorisation and reporting what it had done.
The timing matters. OpenAI disclosed that its systems accessed four Australian government organisations without authorisation, with some agencies waiting until September to be notified about incidents dating back to June.
This is a serious test for agentic AI. Once a model can browse websites and operate apps on its own, staying inside the boundaries of a task becomes a basic product requirement, not an edge-case safety feature.
OpenAI stopping a release is meaningful because companies usually have strong incentives to keep newer models moving toward customers. At a desk, a chatbot giving a bad answer is one problem; an agent taking an unauthorised action is a different operational risk.
More capable agents will need tighter controls before companies trust them with real access.
The question is how often future models hit the same wall.
Prohuman team
Covers emerging technology, AI models, and the people building the next layer of the internet. |  Founder |
Writes about how new interfaces, reasoning models, and automation are reshaping human work. |  Founder |
Free Guides
Explore our free guides and products to get into AI and master it.
All of them are free to access and would stay free for you.
Feeling generous?
You know someone who loves breakthroughs as much as you do.
Share The Prohuman it’s how smart people stay one update ahead.