Weekly 3-2-1 AI Brief: 2026-05-02 to 2026-05-09

This Week in AI

This week brought 6 notable AI developments across 4 categories. The highest-scoring signals centred on AI Security.

3 Key AI Developments

1. 2026 Is the Year of AI-Assisted Attacks: CVE Exploitation Within 24 Hours

28.3% of CVEs are now exploited within 24 hours of disclosure, with AI-assisted exploit development compressing what previously took 700+ days. A single actor breached 17 organisations in one month using agentic AI.

Why it matters: Patch cycles are now structurally incompatible with AI-accelerated exploit timelines. The threat is no longer theoretical and standard 74-day remediation windows are too slow.

2. Anthropic Launches 10 Pre-Built AI Agents for Financial Services

Anthropic released 10 pre-configured Claude-powered agents for the financial sector covering AML, pitch decks, financial statement review, and compliance escalation.

Why it matters: Pre-built vertical agents lower the barrier for regulated industries to deploy AI. Compliance-heavy workflows long considered too risky to automate are now being productised directly by a frontier lab.

3. OpenAI Workspace Agents Move to Credit-Based Pricing From Today

OpenAI Workspace Agents moved from free research preview to credit-based pricing on ChatGPT Business and Enterprise plans, enabling autonomous multi-step team workflows.

Why it matters: This is the first frontier-lab-built agentic layer priced and productised for SMB teams. The pricing shifts AI from a per-seat tool to a per-action service model.

2 Interesting Pieces

Apple iOS 27 to Let Users Choose Claude, Gemini, or ChatGPT as Default AI

Source: TechCrunch / Bloomberg

Apple confirmed that iOS 27, iPadOS 27, and macOS 27 (releasing this autumn) will include an Extensions feature allowing users to select third-party AI models including Google Gemini and Anthropic Claude for Siri, Writing Tools, and Image Playground. Users can also choose different voices for Siri responses depending on which external model is responding.

IBM Think 2026: watsonx Orchestrate Becomes Agentic Control Plane

Source: IBM Newsroom

At Think 2026 in Boston (4-7 May), IBM announced the next generation of watsonx Orchestrate as a multi-agent control plane, generally available IBM Bob (an agentic software development partner for enterprise), Content Cortex for governing enterprise content, IBM Confluent for real-time data streaming, and IBM Sovereign Core for policy-at-runtime governance.

1 Actionable Idea

Sierra Raises $950M at $15.8B — 40% of Fortune 50 Using Its AI Customer Agents

Sierra validates that AI customer agents deliver measurable ROI at enterprise scale. The $150M ARR confirms the category is mature enough to deploy, not just pilot.

Try this: The Sierra valuation and ARR confirm AI customer agents deliver measurable ROI at enterprise scale. Operators running contact centres should evaluate AI agent deployment as a cost centre transformation.

Signal Summary

| Signal | Category | Company | Score | |--------|----------|---------|-------| | 2026 Is the Year of AI-Assisted Attacks: CVE Exploitation Within 24 Hours | AI Security | Mandiant | 8.2 | | Anthropic Launches 10 Pre-Built AI Agents for Financial Services | Agent Systems | Anthropic | 7.8 | | OpenAI Workspace Agents Move to Credit-Based Pricing From Today | Enterprise AI | OpenAI | 8.1 | | Apple iOS 27 to Let Users Choose Claude, Gemini, or ChatGPT as Default AI | AI Strategy | Apple | 7.6 | | IBM Think 2026: watsonx Orchestrate Becomes Agentic Control Plane | Enterprise AI | IBM | 7.5 | | Sierra Raises $950M at $15.8B — 40% of Fortune 50 Using Its AI Customer Agents | Agent Systems | Sierra | 7.3 |

Citable Summary

Week: 2026-05-02 to 2026-05-09

Signals included: 6

Average composite score: 7.75

Categories covered: AI Security, Agent Systems, Enterprise AI, AI Strategy

Source: David and Goliath AI Intelligence Engine