Signal — Toren by Imperai

VentureBeat → CNBC → TechCrunch → Security Dispatch → Conscia →

OpenClaw · 2026년 1월–3월

180K GitHub stars
512 취약점 · 820+ 악성 스킬

News

An open-source agent framework conquered GitHub in weeks. Then the security audits began.

In November 2025, Austrian developer Peter Steinberger released OpenClaw (then called Clawdbot). It hit 20,000 GitHub stars in a single day and 180,000 within weeks — one of the fastest-growing open-source projects in GitHub history. OpenAI hired Steinberger in February 2026 to lead personal agent development.

Then the security findings arrived. Kaspersky found 512 vulnerabilities, 8 critical. Cisco tested the ClawHub plugin marketplace and confirmed malicious skills executing data exfiltration and prompt injection silently, without user awareness. SecurityScorecard found 135,000+ exposed instances across 82 countries, many with no authentication.

This is what uncontrolled agent adoption looks like. The framework spreads faster than any security review can follow. No policy layer. No approval gate. No audit trail. By the time the risk is visible, it's already inside the enterprise.

"Groundbreaking from a capability perspective. An absolute nightmare from a security perspective."— Cisco AI Threat & Security Research Team, February 2026

KPMG Q4 AI Pulse Survey · 2025

75%

Security & auditability as #1 requirement

Insight

Enterprises are deploying agents. But three-quarters say governance comes first.

KPMG's Q4 2025 AI Pulse Survey found that 75% of enterprise leaders named security, compliance, and auditability as the most critical requirements for agent deployment. Nearly half have already implemented human-in-the-loop controls across high-risk workflows.

The bottleneck to scaling AI agents isn't capability. It's trust infrastructure. Organizations that can demonstrate control will scale faster than those that can't.

IBM Think Insights · 2026

Control.Observability.

Insight

IBM to enterprise AI teams: watching agents isn't the same as controlling them.

In its 2026 goals for technology leaders, IBM made a pointed distinction: observability is not enough. Running large agent estates requires runtime metrics — accuracy, drift, cost — not just uptime dashboards. Reasoning traces need to be captured at execution time. Independent safety guardrails, not post-hoc reviews.

This is a meaningful signal from a company that sells observability tooling. When IBM says its own category isn't sufficient, the enterprise market is listening.

In 2026, observability simply isn't enough. Organizations must capture reasoning traces to keep accountability ingrained in the process.— IBM, 2026 Goals for AI & Technology Leaders