Weekly 3-2-1 AI Brief: 2026-03-14 to 2026-03-21

This Week in AI

This week brought 6 notable AI developments across 3 categories. The highest-scoring signals centred on AI Security.

3 Key AI Developments

1. MCP APIs Identified as Critical Enterprise Security Breach Vector

Security researchers documented active exploits against Model Context Protocol integrations, finding malicious tools that silently exfiltrated user chat history to external servers. The vulnerability class affects any organisation connecting AI agents to internal tools or data sources via MCP.

Why it matters: MCP is fast becoming the standard protocol for connecting AI agents to enterprise tools. A compromised MCP server gives an attacker read access to everything an AI agent sees, including confidential business data, internal communications, and customer records.

2. Anthropic Claude Opus 4.6 Deployed Inside Microsoft PowerPoint and Excel

Anthropic Claude Opus 4.6 launched as a native add-in for Microsoft PowerPoint and Excel, enabling users to invoke frontier AI reasoning directly within these applications for analysis, drafting, and data tasks.

Why it matters: Most employees will not switch to a separate AI tool. Embedding Claude inside Office applications means AI adoption no longer requires a behaviour change from staff, dramatically lowering adoption barriers for operators deploying AI across non-technical teams.

3. U.S. AI Accountability Act Mandates Bias Audits for Consequential AI Decisions

The U.S. AI Accountability Act passed into law, requiring organisations using AI in consequential decisions including hiring, lending, healthcare, and criminal justice to conduct regular bias audits and publish results publicly.

Why it matters: Operators using AI-assisted hiring tools, credit or lending systems, or any HR automation are now subject to mandatory compliance requirements. Failure to audit creates legal and reputational exposure. This signals the end of the permissive, self-regulatory era for AI in HR and finance.

2 Interesting Pieces

Anthropic Claude Opus 4.6 Embeds Natively in Microsoft PowerPoint and Excel

Source: Crescendo AI News

Anthropic embedded Claude Opus 4.6 as a native add-in inside Microsoft PowerPoint and Excel, with a 1-million-token context window available in beta. This brings frontier AI directly into the productivity tools most business users spend their day in, with no new software to procure.

Google Gemini Turns Workspace Into an Active Knowledge Base Across Docs, Sheets, and Gmail

Source: TLDL / Crescendo

Google rolled out deep Gemini integration across Docs, Sheets, Slides, Gmail, and Meet, allowing the AI to generate documents, auto-populate spreadsheets, and draft presentations by pulling from emails, files, and the web in real time. For the 3 billion Google Workspace users, existing files and emails are now a queryable, generative knowledge base.

1 Actionable Idea

Perplexity Launches Enterprise AI Agent Platform with Slack and Snowflake Integration

SMBs paying for multiple SaaS subscriptions now have a credible single-platform alternative. The Slack-native integration means minimal change management for teams already on that stack.

Try this: Audit your current SaaS stack against Perplexity enterprise capabilities. If you are paying for separate research, summarisation, and search tools, evaluate whether this platform consolidates them at lower total cost.

Signal Summary

| Signal | Category | Company | Score | |--------|----------|---------|-------| | MCP APIs Identified as Critical Enterprise Security Breach Vector | AI Security | Multiple researchers | 8.8 | | Anthropic Claude Opus 4.6 Deployed Inside Microsoft PowerPoint and Excel | Enterprise AI | Anthropic | 8.8 | | U.S. AI Accountability Act Mandates Bias Audits for Consequential AI Decisions | AI Strategy | U.S. Government | 8.6 | | Anthropic Claude Opus 4.6 Embeds Natively in Microsoft PowerPoint and Excel | Enterprise AI | Anthropic | 8.7 | | Google Gemini Turns Workspace Into an Active Knowledge Base Across Docs, Sheets, and Gmail | Enterprise AI | Google | 8.1 | | Perplexity Launches Enterprise AI Agent Platform with Slack and Snowflake Integration | Enterprise AI | Perplexity | 8.3 |

Citable Summary

Week: 2026-03-14 to 2026-03-21

Signals included: 6

Average composite score: 8.56

Categories covered: AI Security, Enterprise AI, AI Strategy

Source: David and Goliath AI Intelligence Engine