Insurance companies carve out exceptions for AI risks: U.S. military re-engages with Anthropic

In today’s edition of Data Points, you’ll learn more about:

• A San Francisco retail store run by an AI agent
• OpenAI’s specialized life-sciences model
• Anthropic’s designs on the market for graphic design
• Wall Street’s AI-driven workforce adjustments

But first:

Insurers move to limit their exposure to AI-related claims

Major insurers including units of Berkshire Hathaway, Travelers Group, and Chubb Limited are seeking to exclude or restrict coverage in standard liability policies of damage caused by artificial intelligence systems. The shift reflects concerns that AI-driven errors such as faulty outputs, fraud enabled by deepfakes, or failures of automated decision-making could generate large, hard-to-model claims, leading insurers to exclude such events or require new, AI-specific coverage. As insurers narrow their coverage, companies that deploy AI systems may need to absorb risk directly or purchase specialized policies. (Futubull)

U.S. officials reopen talks with Anthropic despite recent dispute

Anthropic continues discussions with senior members of the Trump administration despite an earlier Pentagon decision to label the company a supply-chain risk and restrict its use in defense contexts. The engagement includes briefings and meetings around Anthropic’s Claude Mythos model. Ongoing talks between the company and U.S. officials suggest that government access to advanced AI systems may outweigh prior disagreements. (TechCrunch)

AI-run retail experiment exposes limits of autonomous agents

An AI startup, Andon Labs, opened a retail store in San Francisco that is designed and managed by an AI agent named Luna. The agent received budget, tools, and operational goals and handles tasks such as selecting inventory, negotiating with vendors, hiring workers, and interacting with customers. However, The New York Times report highlights frequent errors in judgment. The deployment exposes the gap between current agent capabilities and business requirements, providing a concrete test bed for evaluating reliability, oversight, and failure modes in agentic systems. (The New York Times)

OpenAI fine-tunes AI model for life sciences workflows

OpenAI introduced GPT-Rosalind, a large language model trained on biology workflows and made available through a restricted access program. The model differs from general-purpose models by targeting domain-specific processes used in life sciences. It’s fine-tuned for tasks such as querying scientific databases, synthesizing research, generating hypotheses, and planning experiments. GPT-Rosalind emphasizes the potential value of specialized systems that reduce reliance on general LLMs for work in highly differentiated domains. (Ars Technica)

Anthropic expands into design workflows

Anthropic introduced Claude Design, an experimental product that generates visual assets including slide decks, logos, and leaflets from natural language prompts. The tool targets non-designers such as founders and product managers. It produces editable drafts that users can refine within a conversational interface. Claude Design illustrates how multimodal models can absorb parts of the product design workflow traditionally handled by specialized tools and teams. (TechCrunch)

AI drives job cuts across Wall Street banks

Wall Street banks including Goldman Sachs and Morgan Stanley are restructuring roles like research, trading support, and back-office operations as they deploy AI systems. The firms are using AI for tasks such as financial modeling, drafting reports, and processing large datasets, which reduces demand for junior analysts and operational staff while increasing demand in technical infrastructure. As firms integrate AI into core workflows, the financial industry is using smaller, more-technical teams, potentially affecting hiring patterns and career paths in a lucrative sector for engineering talent. (The New York Times)

Want to know more about what matters in AI right now?

Read the latest issue of The Batch for in-depth analysis of news and research.

Last week, Andrew Ng talked about how coding agents accelerate software work to varying degrees, with the most impact on frontend development, followed by backend, infrastructure, and research, and how this understanding helps in organizing software teams effectively.

“Backend development — say, building APIs to respond to queries requesting product data — is harder. It takes more work by human developers to steer modern models to think through corner cases that might lead to subtle bugs or security flaws. Further, a backend bug can lead to non-intuitive downstream effects like a corrupted database that occasionally returns incorrect results, which can be harder to debug than a typical frontend bug.”

Read Andrew’s letter here.

Other top AI news and research covered in depth:

• GLM 5.1 Aims for Long-Running Tasks as Z.ai’s latest model evaluates interim results and changes its approach hundreds of times before delivering final output.
• Humanoid Robots Work Factory Floors with Agility Digits robots fetching and carrying bins at a Schaeffler auto-parts factory, displacing humans into higher-level jobs.
• Anti-Data-Center Revolt Gains Traction as public opposition to new data centers in the U.S. spurs political action and violence.
• Assistants That Assisted Consistently addresses the drift of large language models from helpful to harmful personas, with new research aiming to stabilize them.

Last chance!
A special event for our community

Andrew Ng and DeepLearning.AI are hosting AI Dev 26 × San Francisco, a two-day conference for AI developers taking place April 28-29 at Pier 48.

Join 3,000+ engineers, researchers, and builders working on modern AI systems.

The program includes top speakers, developer relations experts, and engineers from companies including Google, AMD, Oracle, Neo4j, and Snowflake (and of course DeepLearning.AI), all sharing their latest technologies and explaining how they’re building and deploying AI systems today.

At AI Dev 26, you’ll find:

• Technical talks from engineers building AI systems in production
• Hands-on workshops exploring new tools and techniques
• Live demos from startups and AI builders
• Opportunities to meet other developers and companies in the space

Get your ticket with a special discount!

Data Points is produced by human editors with AI assistance.