<?xml version="1.0" encoding="utf-8"?><rss version="2.0"><channel><title>ctxsec: Context &amp; Harness Attacks</title><link>https://ctxsec.com</link><description>context security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.</description><item><title>Defender for Office 365: prompt-injection protection quarantines high-confidence email before Copilot sees it</title><link>https://ctxsec.com/context#mdo-prompt-injection-protection</link><guid>https://ctxsec.com/context#mdo-prompt-injection-protection</guid><pubDate>Tue, 13 Oct 2026 00:00:00 +0000</pubDate><category>confirmed (vendor)</category><description>Microsoft documents a mail-flow layer that detects prompt-injection content in inbound email before delivery to users or AI assistants. Detections land as High confidence phishing with technology &quot;Prompt injection protection&quot; (MDO Plan 2 / E5-class). Focus: instructions that try to exfiltrate via URL, reveal system prompts, or enumerate write-capable tools — including hidden/white-on-white text. Runtime Copilot safeguards remain a second layer. This is harness/context defense at the email channel, not a new CVE.</description></item><item><title>State of AI Cybersecurity: harness-layer attack proposals up 175% (12 → 33), March to October</title><link>https://ctxsec.com/context#unprompted-state-of-ai-cyber</link><guid>https://ctxsec.com/context#unprompted-state-of-ai-cyber</guid><pubDate>Tue, 13 Oct 2026 00:00:00 +0000</pubDate><category>confirmed (author&#x27;s post)</category><description>Rob T. Lee and Indre Altman compared the [un]prompted AI Security Practitioner Conference CFP rounds: 469 submissions in March and 490 in October. Harness-layer attack proposals rose from 12 to 33. 94 submissions used the word &#x27;harness&#x27;. 71% of the October pool is agent-native (harness, MCP and skill supply chain, agent identity, prompt injection, agentic-browser attacks). Only 12% carry production evidence. Caveat: this is a CFP census, not incident data. The full report is linked from the conference site; we&#x27;ve only verified the figures quoted in the author&#x27;s post.</description></item><item><title>SafeHarness: defenses built into the agent harness lifecycle</title><link>https://ctxsec.com/context#safeharness</link><guid>https://ctxsec.com/context#safeharness</guid><pubDate>Wed, 15 Apr 2026 00:00:00 +0000</pubDate><category>research (benchmark, not production)</category><description>An arXiv paper treating the harness (tool use, context management, state) as the attack surface. It proposes four defense layers: adversarial context filtering, causal verification of decisions, privilege-separated tool control and safe rollback. It reports ~42% lower attack success on benchmarks. Caveat: simulated evaluation, not production evidence.</description></item></channel></rss>