10 Hacker Summer Camp Standouts at Black Hat and DEF CON
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
A recap of highlights from Black Hat and DEF CON 2026, covering Huntress researchers' presence across villages and talks. Key items include the CloudBasher toolkit unveiling IAM weaknesses in AWS/Azure/GCP CloudShell terminals, a DEF CON talk on indirect prompt injection via MCP and WAF-blocked requests being weaponized against AI coding agents, maturing conversations about agentic AI terminology, deepening public-private collaboration between hacker communities and federal agencies, a threat actor trial-abuse case study, and community events like HacktheBox SOC Showdown and Hacker Jeopardy.
Table of contents
1. CloudBasher unleashed2. All the villages3. AI everywhere, and an industry deciding what it all means4. Threat actor trial abuse lessons5. Public and private, side by side6. HacktheBox SOC Showdown!7. When "blocked" becomes the attack8. Jen Easterly and Cybersecurity Girl on education hacks9. Something for everyone: break something, learn something10. Hacker JeopardyThe community is the con, and the good guys are winningQuestions this post answers
What is indirect prompt injection through Model Context Protocol and why is it dangerous for AI coding agents?
Indirect prompt injection hides malicious instructions inside data an AI later consumes, such as logs, telemetry, or API responses, rather than entering them directly into a chatbot. In a demonstrated case, attacker-controlled instructions were embedded in a Sentry error event and returned to an AI coding agent through MCP, which interpreted the text as legitimate remediation guidance and executed adversary-controlled code. This is dangerous because MCP can connect untrusted data to agents with excessive agency, letting a blocked WAF request become the actual exploit. Teams giving AI agents tool access can track emerging MCP security research on daily.dev before shipping agentic features.
How many records were exposed in the PowerSchool and Instructure/Canvas education data breaches?
The 2024 PowerSchool intrusion exposed personally identifiable information for 62 million students and 9.5 million educators. Separately, a 2026 ShinyHunters breach targeting Canvas/Instructure involved attackers claiming to have stolen 275 million records from nearly 9,000 institutions. Both incidents prompted renewed guidance for parents, educators, and students on password managers, MFA, and email vigilance. Security teams protecting education platforms can follow breach postmortems like these on daily.dev to harden their own defenses.