AI Roundup: April 2, 2026

1. Q1 2026 Startup Funding Shatters All Records at $297 Billion

Crunchbase data shows investors poured $297 billion into startups globally in Q1 2026, a 2.5x increase over the previous quarter’s $118 billion and more than many full-year totals from prior years. Four of the five largest venture rounds ever recorded closed in the quarter — OpenAI ($122B), Anthropic ($30B), xAI, and Waymo — collectively accounting for $188 billion or 63% of the total. The concentration in AI mega-rounds underscores how a handful of U.S.-based companies are absorbing a disproportionate share of global venture capital. Source

2. Salesforce Transforms Slackbot into Enterprise AI Agent with 30 New Features

Salesforce launched more than 30 new AI capabilities for Slackbot in its most sweeping update since the $27.7 billion Slack acquisition in 2021. The overhaul turns Slackbot into a full-spectrum enterprise agent that takes meeting notes across video providers, operates on users’ desktops outside Slack, and executes tasks through third-party tools via the Model Context Protocol (MCP). Slack also introduced reusable AI-Skills — instruction sets that teams build once and deploy on demand — and integrated Slackbot as an MCP client for Agentforce, Salesforce’s agent development platform. Source

3. Oracle Cuts Thousands of Jobs While Ramping AI Infrastructure Spending

Oracle began notifying employees of a sweeping layoff that could affect 20,000 to 30,000 workers from its 162,000-strong workforce. TD Cowen analysts estimate the cuts could yield $8-10 billion in incremental free cash flow. The restructuring, with costs projected at up to $2.1 billion, comes as Oracle accelerates data center buildout for AI workloads following a February announcement to raise up to $50 billion in debt and equity for cloud capacity contracted by NVIDIA, Meta, OpenAI, and xAI. Oracle shares rose 6% on the news despite being down 25% year-to-date. Source

4. Tenex Raises $250M for AI-Powered Cybersecurity, Reaches Unicorn Status

Tenex, a Sarasota-based cybersecurity startup, raised $250 million in Series B funding led by Crosspoint Capital with participation from Shield Capital and DeepWork Capital, valuing the company above $1 billion. Previously backed by Andreessen Horowitz with a $27 million Series A in September 2025, Tenex offers a managed detection and response service built entirely on AI agents that escalate to humans when necessary. The company plans to hire more than 250 engineers and salespeople to support global expansion amid accelerating enterprise demand for AI-driven security operations. Source

5. Anthropic Blames Claude Code Leak on “Process Errors,” Scales Back DMCA Takedowns

Anthropic disclosed that the accidental exposure of Claude Code’s source code via npm was caused by “process errors” in the company’s rapid product release cycle, not a security breach. After initially filing broad copyright takedown requests affecting thousands of GitHub repositories, Anthropic narrowed its DMCA claims to 96 specific copies and adaptations, acknowledging the original scope was unintentionally wide. Meanwhile, developers circumvented some removals by rewriting the exposed code in different programming languages, ensuring the leaked implementation details remain widely circulated despite the legal effort. Source

6. Realbotix Plans Delivery of 19 Humanoid Robots Across Spring 2026

Realbotix announced it will deliver 19 humanoid robots and corresponding AI systems across March, April, and May 2026 as it scales production for homes, healthcare, hospitality, and customer-facing roles. Each unit features eye-tracking, AI vision, conversational intelligence with memory, and a 10-hour battery life with continuous operation capability when plugged in. The deliveries represent an early test of consumer and enterprise appetite for general-purpose humanoid robots outside industrial settings. Source

7. Elgato Stream Deck Adds MCP Support, Letting AI Assistants Trigger Macros

Elgato released Stream Deck software version 7.4 with built-in Model Context Protocol (MCP) support, enabling AI assistants including Claude, ChatGPT, and Nvidia G-Assist to find and activate Stream Deck actions on a user’s behalf. The update makes the Stream Deck function as an MCP server, so connected AI tools can discover and trigger any configured macro without the user pressing a physical button. MCP — backed by Anthropic, Microsoft, Figma, and Canva — is becoming the standard integration layer between AI assistants and desktop applications, and this marks one of its first consumer hardware integrations. Source

8. Baidu Apollo Robotaxis Freeze in Wuhan, Trapping Passengers and Snarling Traffic

Baidu’s Apollo Go robotaxi service experienced a mass system failure in Wuhan on April 1, with police confirming reports of vehicles stopping in the middle of streets and being unable to move, reportedly trapping passengers inside and causing at least one accident. Preliminary investigations attributed the incident to an unspecified “system failure”; Baidu, which has deployed more than 500 driverless vehicles in Wuhan alone and operates in 26 cities worldwide including London and Dubai via Uber, did not immediately comment. Local news cited by Reuters estimated at least 100 vehicles were affected, reigniting safety debates around autonomous vehicle deployment in dense urban environments. Source

9. ChatGPT Arrives in Apple CarPlay with iOS 26.4

OpenAI’s ChatGPT is now accessible directly from the CarPlay dashboard for users running iOS 26.4 or newer with the latest ChatGPT app. Apple’s iOS 26.4 update introduced support for voice-based conversational apps in CarPlay, allowing AI chatbots to integrate into the in-car platform; per Apple’s developer guidelines, the CarPlay interface shows no text responses — all interaction is voice-only, with on-screen buttons limited to mute and end conversation. Users can also review a list of recent conversations from the CarPlay interface, though a wake word is not supported and the app must be tapped to open. Source

10. Gig Workers Are Training Humanoid Robots at Home

Physical Intelligence, Figure, and other humanoid robotics startups are recruiting gig workers to perform household task demonstrations at home, recording physical interactions via webcam and motion-capture hardware to build embodied AI training datasets. MIT Technology Review reports that contractors earn $15-25 per hour performing repetitive actions designed to teach robots manipulation skills, mirroring the data-labeling gig economy that powered early large language model development. The practice has prompted debate about whether human-demonstration data can scale fast enough to meet the training demands of general-purpose robots, with some researchers arguing synthetic data generation will be necessary to hit commercial deployment targets. Source

11. Researchers Call for AI Benchmark Overhaul as Top Models Saturate Existing Tests

AI researchers and evaluation teams are pushing for a fundamental redesign of AI benchmarks, with MIT Technology Review publishing analysis showing that leading models now saturate tests including MMLU, HumanEval, and GSM8K before those scores reliably predict real-world deployment performance. The authors argue the field needs benchmarks anchored to specific task domains — legal reasoning, medical diagnosis, software engineering — with clear human baselines and reproducible evaluation protocols. Several major labs have already announced in-house red-teaming evaluation programs, in some cases replacing third-party benchmark comparisons in their public model release notes. Source

12. Mercor Breached via LiteLLM Open-Source Supply Chain Attack

Mercor, an AI-powered technical recruiting platform, confirmed a cyberattack after the Lapsus$ extortion group claimed responsibility for stealing data from the company’s systems by exploiting a compromise of the open-source LiteLLM project. LiteLLM is a widely used library for routing requests across multiple LLM providers and is integrated into hundreds of AI applications and developer tools, making any compromise a supply-chain risk for the broader AI ecosystem. Mercor said it was working to notify affected parties; the incident follows a pattern of attackers targeting open-source AI infrastructure projects as an entry vector into the companies that depend on them. Source