Anthropic Releases Claude Opus 4.6 with Agent Teams and PowerPoint Integration
Anthropic dropped Opus 4.6 on February 5 — and the headline capability isn't the benchmark numbers. It's agent teams: multiple Claude instances collaborating on complex, multi-step tasks, with one acting as an orchestrator. The model also integrates directly into PowerPoint. TechCrunch confirmed this is particularly significant for Claude Code users, where Opus 4.6 now orchestrates multi-agent coding workflows without human handoffs between steps. The CNBC framing — "vibe working" — is cute but undersells it. If you're running Claude in enterprise workflows, agent teams are the architectural shift that changes what's possible. Run one pilot in the next 30 days before your competitors do it for you.
OpenAI Launches Lockdown Mode for High-Security Enterprise Users
OpenAI introduced Lockdown Mode — an optional, aggressive security setting that deterministically disables any ChatGPT capability that could exfiltrate data via prompt injection. In Lockdown Mode, web browsing is limited to cached content; no live network requests leave OpenAI's infrastructure. It ships alongside "Elevated Risk" labels that flag specific capabilities — like Codex's network access — that security teams should evaluate before enabling. This is the most explicit acknowledgment yet from OpenAI that prompt injection is a real enterprise threat, not a theoretical one. If your security team hasn't briefed your AI governance committee on prompt injection risks, Lockdown Mode's existence is your slide deck.
Experian 2026 Forecast: AI Agents Are the Emerging Top Data Breach Vector
Experian's 2026 data breach forecast, released in late January, put a number on something practitioners already knew: AI agents are becoming the preferred attack surface, poised to replace human error as the leading cause of breaches. Hackers can now inject their own agent instructions into your agents via prompt injection, causing them to exfiltrate data at machine speed. The attack doesn't need a phishing email. It needs a poorly sandboxed agent with write access. Exabeam launched an AI Agent Security monitoring product to address exactly this; Experian's forecast tells you why the market for it will be large. If your agents have access to sensitive data, that access needs least-privilege controls today.
→ Read more at www.thesignal.press