This week, AI drove a rover across Mars, three competing labs declared a joint war on Chinese AI theft, and Oracle fired 30,000 people to fund the infrastructure that makes all of it possible. The ‘AI is hype’ debate is over.
THIS WEEK’S TOP 3 MUST-KNOW CONCEPTS
Anthropic’s Mythos: The AI Too Dangerous to Release
AI SAFETY
Anthropic unveiled Claude Mythos this week, a frontier model so capable it reportedly found critical vulnerabilities in every major OS and web browser, including one undetected for 27 years. Rather than a public launch, Anthropic is running Project Glasswing: a controlled pilot with Goldman Sachs, Citigroup, and JPMorgan to use Mythos defensively before bad actors find their own access. OpenAI is already building a rival capability in response. For SME leaders, this signals that AI has crossed into nation-state-level security intelligence. The practical implication: your next cybersecurity vendor may well be a model, not a firm, and the window to build those relationships is now. OpenAI, Google and Anthropic Unite Against Chinese AI Theft
OpenAI, Google and Anthropic Unite Against Chinese AI Theft
GEOPOLITICS
Three competing AI labs are now sharing threat intelligence through the Frontier Model Forum to counter adversarial distillation by Chinese firms. Anthropic alone documented 16 million unauthorised exchanges from DeepSeek, Moonshot AI, and MiniMax via roughly 24,000 fraudulent accounts. The real concern goes beyond stolen IP: distilled copies do not inherit safety filters, producing unguarded frontier models that could be deployed for surveillance or disinformation at government scale. US officials estimate the practice costs American AI labs billions annually. This is the first coordinated cross-competitor defence operation in AI history, and it signals that the competitive dynamics of the industry are shifting fast.
Oracle Cuts 30,000 Jobs to Fund AI Infrastructure
WORKFORCE
Oracle executed the largest layoff in its history on 31 March, cutting up to 30,000 employees (18% of its workforce) via a 6am email and redirecting the estimated $8-10 billion in annual savings to AI data centre infrastructure. Oracle is not alone: more than 103,000 tech workers lost roles in Q1 2026, with AI explicitly cited in at least 20% of those cuts. Companies like Block, Atlassian, and Dell have directly attributed workforce reductions to AI-driven efficiency gains. For SME executives, the ‘AI creates as many jobs as it destroys’ thesis is now in its stress-test phase. Workforce planning conversations that felt optional 12 months ago are now urgent.
OTHER MAJOR STORIES THIS WEEK
Claude Plans and Executes the First AI-Driven Mars Rover Route
SPACE + AI
In December 2025, NASA’s Perseverance rover completed the first drives on another world planned entirely by AI. Claude analysed orbital imagery, identified hazards, and generated 456 metres of safe waypoints validated against 500,000 telemetry variables before transmission. Engineers estimate this cuts route-planning time in half. As signal delays to deeper space stretch from minutes to hours, autonomous AI navigation shifts from useful to essential.
Salesforce Turns Slack into a Full Agentic Interface
PRODUCTIVITY
CEO Marc Benioff personally presented a major Slack overhaul, transforming it from a messaging tool into an agentic execution layer. The revamped Slackbot can retrieve budgets, adjust figures based on new inputs, and orchestrate actions across Salesforce CRM in natural language. This is a direct competitive challenge to Microsoft Copilot Cowork. For SME leaders whose teams already live in Slack, this is worth evaluating now.
AI Self-Improvement Accelerates: Anthropic Claims 90% Code Authorship
AI DEVELOPMENT
Anthropic states that Claude now authors up to 90% of code on some internal projects; OpenAI plans to deploy an AI “intern” to its own research workflows within six months. MCP, the open standard for agent-to-tool connectivity, crossed 97 million installs in March 2026. The pace of capability improvement is widening faster than most business planning cycles anticipate.
QUICK DIGEST
Amazon Ring Adds Facial Recognition
PRIVACY
Amazon’s Familiar Faces feature identifies visitors via Ring cameras using cloud AI. Consumer groups and lawmakers have raised alarm over surveillance risks and stricter biometric laws in three US states have already blocked it. For SME leaders, this previews the regulatory friction AI identity tools will generate.
Meta Enters the Reasoning Race with Muse Spark
AI MODELS
Meta released Muse Spark, a reasoning-focused model targeting the gap with the top three labs. With Meta’s reach across Facebook, Instagram, and WhatsApp, a credible reasoning model could alter competitive dynamics considerably over the next two quarters.
OpenAI Eyes Late 2026 IPO at $25B Revenue
FINANCE
OpenAI has crossed $25 billion in annualised revenue and is taking early steps toward a public listing. Rival Anthropic is approaching $19 billion. For business leaders assessing AI vendors, the financial stability of your provider is now a legitimate due-diligence question.
AI Caused 25% of US Job Cuts in March 2026
WORKFORCE
New data shows AI was explicitly cited in 25% of all US job cuts in March 2026, roughly 15,341 roles in a single month. The roles most affected are data analysis, back-office operations, and customer service. This should inform every SME workforce planning conversation right now.
Google TurboQuant Cuts AI Memory Costs Sixfold
EFFICIENCY
Google Research unveiled TurboQuant at ICLR 2026, reducing memory overhead of running large AI models by a factor of six. The practical result is faster, cheaper AI inference. As these gains flow through to API pricing, the cost barrier for SME AI adoption continues to fall.
Agentic AI Foundation Formed Under Linux Foundation
OPEN SOURCE
The Linux Foundation has established the Agentic AI Foundation, anchored by Anthropic’s MCP, OpenAI’s AGENTS.md, and Block’s Goose framework. With MCP at 97 million installs, this open standard is becoming the default plumbing for agent-to-tool connectivity. Build on MCP-compatible infrastructure.