AI News: April 24, 2026

1. Sierra Makes Its Third Acquisition of 2026, Buying YC-Backed Fragment

Sierra. Bret Taylor’s customer-service agent company Sierra acquired French YC-backed startup Fragment, which builds tooling to integrate AI into business workflows. Terms were not disclosed; Fragment had raised roughly $2M in seed. Fragment’s founders Olivier Moindrot and Guillaume Genthial are joining Sierra, and Taylor framed the deal as strengthening Sierra’s “agent development efforts in France.” This is Sierra’s third public acquisition of the year after Japan’s Opera Tech and voice-agent company Receptive AI, continuing a rapid consolidation pattern from a startup that is now reportedly valued at $10 billion. Source

2. Noscroll Launches an AI Bot That Doomscrolls So You Don’t Have To

Noscroll. Former OpenSea CTO Nadav Hollander and the developer known as @z0age launched Noscroll, a $9.99/mo service that crawls X, news sites, Reddit, Hacker News, and Substack on a user’s behalf and texts back AI-summarized digests at the user’s preferred cadence. It is a small but symptomatic product: a whole consumer category — “read the internet for me” — is now viable at a price that would have been impossible without cheap frontier-model inference. Source

3. Era Raises $11M to Build a Software Platform for AI Gadgets

Era. Era, founded by ex-Humane CEO Liz Dorman, closed a combined $11M ($9M seed led by Abstract Ventures and BoxGroup, plus a $2M pre-seed from Topology Ventures and Betaworks), with angels including Flickr’s Caterina Fake and iPhone-keyboard designer Ken Kocienda. Era is not building its own device — it is building the software layer (130+ LLMs, 14+ providers) for other hardware makers shipping glasses, rings, and pendants. The bet is that the Humane/Rabbit-style “AI device” category is still going to happen, just not with one company owning it end-to-end. Source

4. Another Delve Customer Gets Breached, This Time Context AI via Vercel

Delve. TechCrunch reported a new breach implicating compliance-certification startup Delve: Context AI, which held Delve-issued certifications, had an employee connect a Context AI application to Vercel’s corporate Google account — and attackers used that connection to break into Vercel’s internal systems and exfiltrate customer data. It is the latest in a series of incidents (LiteLLM, Lovable) around Delve customers, and it comes after Delve already lost Y Combinator’s backing amid accusations of faked customer data and rubber-stamping auditors. Source

5. AI-Driven Astronomy Is Now Big Enough to Contribute to the GPU Crunch

Research. A TechCrunch feature walks through how space-telescope data has exploded — Nancy Grace Roman will deliver 20,000 TB over its lifetime, Webb produces 57 GB daily versus Hubble’s 1-2 GB, and Vera C. Rubin will collect 20 TB nightly — and how astronomy groups are competing with industry for GPUs to run AI analyses. UC Santa Cruz’s Brant Robertson notes that university GPU clusters are aging out while demand grows, a problem made worse by a proposed 50% NSF budget cut. It is a useful data point on who is actually buying the Hoppers and Blackwells that aren’t going to frontier labs. Source

6. Sony AI Says Ace Is the First Robot to Reach Expert Level in a Sport

Sony. Sony AI claims its table-tennis robot “Ace” is the first robot to hit expert-level human play in a sport. In matches staged through late 2025, Ace defeated three of five elite amateur players and, by December, was beating professionals. The system uses nine cameras to track ball spin faster than humans can perceive and runs on AI-powered control. It is the most concrete recent data point in the ongoing embodied-AI vs. pure language-model debate. Source

7. Researchers Warn Washington Is Repeating Its ChatGPT Mistake, This Time with World Models

Policy. Stanford HAI’s Russell Wald and Tulane’s Blaine Fisher warn that US policymakers are broadly unaware of world models — AI systems that predict physical environments rather than text — and risk the same under-reaction that preceded ChatGPT. The argument: world models bring distinct risks around privacy, labor, and national security (enhanced surveillance, autonomous weapons), demand massive compute plus physical hardware, and are already being advanced aggressively by China. The piece comes the same week a Honor robot broke the human half-marathon record. Source

8. Trump Science Advisor Says Chinese Actors Are Copying US Frontier Models at Industrial Scale

Policy. The US government’s science advisor told reporters that Chinese actors are carrying out large-scale industrial distillation of American frontier models — essentially using American APIs to train copies — and named China as the primary culprit. The framing suggests upcoming policy moves around API access controls, though nothing concrete has been announced. Whether the claim holds up technically, it sets up the rhetorical backdrop for any export-control or API-residency actions that follow. Source