Salesforce and the Deutscher Mittelstands-Bund surveyed 700 mid-market companies for the KI Index Mittelstand 2026. The headline: 51.2% now use or test AI, up 54% from 33.1% in 2024. AI agent adoption nearly doubled to 16.6%. But the detailed numbers reveal a Mittelstand splitting in two, with knowledge gaps, data protection fears, and legal uncertainty keeping 40% of companies on the sidelines.
E2B caps your sandbox sessions at 24 hours. Modal locks you into their cloud. Daytona runs on AGPL-3.0, which poisons your codebase if you embed it. If you are building AI agents that execute untrusted code, these constraints matter more than boot time benchmarks. Two new open-source tools, both written in Rust and licensed under Apache 2.0, aim to fix this: Microsandbox (5,100+ stars) gives you programmable MicroVM sandboxes with network interception and a plugin system. BoxLite (1,600+ stars) strips sandboxing down to an embeddable library with no daemon, no cloud account, and sub-50ms boot times. ...
Radware launched the industry’s first agentic security posture management solution on February 3, 2026, weeks after discovering the ZombieAgent zero-click vulnerability. The product monitors AI agents at runtime using behavioral analysis, maps multi-agent risk paths, and integrates with Microsoft Copilot, AWS Bedrock, and custom-built agents. This post breaks down the four pillars, what the product actually does, and where it fits in the emerging AI agent security market.
BodySnatcher (CVE-2025-12420, CVSS 9.3) chained a hardcoded platform secret with email-based auto-linking to let attackers impersonate any ServiceNow user, including admins, and issue natural language commands to AI agents. With 85% of Fortune 500 companies running ServiceNow, this was the most severe agentic AI vulnerability disclosed to date. Here is exactly how the attack chain worked and what it means for securing AI agents in enterprise SaaS.
The International AI Safety Report 2026 documented that AI agent capabilities are outpacing safety frameworks. In response, DFKI CEO Antonio Krüger called for a dedicated German AI Safety Institute to advise the federal government. Germany and Italy are the only major AI nations without one. This post covers what a German AI Safety Institute would do, why the DLR research institute is not enough, and how it connects to KI-MIG and BSI efforts.
AI was supposed to make recruiting faster and cheaper. Robert Half’s March 2026 survey of 2,000 hiring managers shows the opposite: 67% say AI-generated applications have slowed hiring, 84% report heavier workloads, and average cost-per-hire has climbed to $4,800. This post quantifies the real cost of the AI recruiting arms race for both employers and candidates, and identifies the three changes that actually reduce spending.
Thirty-eight researchers from Harvard, MIT, Stanford, CMU, and Northeastern gave six AI agents real email accounts, shell execution, Discord access, and 20GB of file storage. Over two weeks, the agents leaked secrets, destroyed their own email servers, obeyed imposters, and propagated unsafe behavior to each other. None of it required adversarial prompting. These failures emerged from normal operation.
The next breach at your company probably won’t come from a phished employee or a stolen laptop. It will come from an AI agent you deployed yourself. That is the core warning from Palo Alto Networks Chief Security Intelligence Officer Wendi Whitmore, who named AI agents the single biggest insider threat of 2026. Not a theoretical risk. Not a future concern. The threat vector that CISOs should be losing sleep over right now. ...
Snyk open-sourced agent-scan, a CLI tool that auto-discovers agent configurations and scans MCP servers and agent skills for 15+ risk categories including tool poisoning, prompt injection, and credential theft. Backed by the Invariant Labs acquisition and the ToxicSkills report that found security flaws in 36.8% of skills, agent-scan is the first dedicated security scanner for AI agent supply chains.
Anthropic launched Claude Computer Use on March 24, 2026. You text Claude a task from your iPhone, and it completes the work on your Mac by controlling the screen, mouse, and keyboard. Here is how the three-tier priority system works, what Dispatch adds, and why security researchers are already raising flags.