Kevin Backhouse

🎺 🎺 Have *you* developed applications with client-side secrets? It can be a real pain since secret management APIs are different on all platforms. We have released MPSS---a Multi-Platform Secure Signing library. It provides a unified secret management API that is backed by native HSM-powered secret management APIs on different platforms (Windows, macOS, iOS, Android). It also comes with an OpenSSL provider that allows you to use MPSS secrets through the familiar OpenSSL APIs that we all know and love. MPSS is available at https://lnkd.in/ghQ6DtTY. We welcome code contributions from the community and hope you find the library useful. Huge thanks to my amazing collaborator Radames Cruz for making this real.

54

I came across a surprising GitHub earlier today 🤖 https://lnkd.in/eZkgXbr5 A frustrated site owner reported that his site was being scraped aggressively with Scrapy, a popular open-source scraping framework. His proposed fixes? – Enforce robots.txt at the framework level – Limit concurrency by default – Require users to insert an email in the user agent If that sounds like trying to stop a freight train with a “Please slow down” sign, that’s because it is 😅 . The maintainers gave the expected response: Scrapy is just a tool. They’re not responsible for how it’s used. If you're seeing abusive traffic, you need to defend your site, not ask bot developers to be nicer 😇 Still, this thread is a great example of how many teams encounter the bot problem for the first time: by surprise, and completely unprepared. They assume bots will follow the rules. That "good behavior" will be enough. That robots.txt is a real deterrent. It’s not. The reality? If your site has anything of value—inventory, PII, login endpoints, limited edition drops, pricing APIs—sooner or later, fraudsters and bot operators will come after it. Scrapers might be the first wave. Next comes credential stuffing, fake account creation, card testing, or abuse of free trials. And they won’t politely identify themselves. They’ll spoof user agents. rotate IPs. Use headless browsers, anti-detect tools, residential proxies. Basic rate limits and blocklists won’t cut it. Hope is not a strategy. Neither is trusting that user agents are honest. If you’re not actively investing in bot detection, you’re not protected. You’re just lucky, for now 🤞

48

2 Comments

🚨 A single misconfigured debug port can put your entire application at risk. Our latest blog post examines the real-world risks of exposed Java Debug Wire Protocol (JDWP) endpoints. In some cases, attackers began exploiting open -JDWP ports within hours of exposure: - RCE via open JDWP ports - Deployment of cryptominers and data exfiltration - Persistence through systemd services, cron jobs and shell startup scripts Read the full analysis: https://lnkd.in/dabGyydZ

128

2 Comments

If you think Prompt Injection is just a "chat" problem, you’re wrong. The Orca Security research team (<3) just proved you can take over a GitHub repository without ever typing a single word to the AI. I'm so excited to see this research project, which Roi Nisimi led before I left, finally see the light of day. No "ignore previous instructions." No "jailbreaking." No social engineering. The attack is just a "comment" in a GitHub Issue. When a developer opened their IDE to help, the AI read that comment, saw it as a set of commands, and silently handed the attackers the repo's GITHUB_TOKEN. This is the "Compound Risk" of Agentic AI. When you give an AI "hands" (tools, terminal access, tokens) and a "brain" that can't tell the difference between a user's instruction and a piece of data it just read, you've built a Sleeper Agent. If your AI agent can read your Jira tickets, your Slack, or your PRs, an attacker doesn't need to hack you. They just need to "salt" the information the AI is about to process. In reality: We are giving AI "God Mode" permissions while it's still operating on "Open Book" logic. That's an architectural disaster waiting to happen. Stop filtering the chat. Start hardening the architecture.

60

10 Comments

UDRL + Sleepmask is pretty powerful... with it you can extend Cobalt Strike's BeaconGate capability to any API you want. The attached graphic demonstrates how HttpOpenRequestA can be forwarded through BeaconGate even though it is not one of the officially supported API's. Another neat feature for the UDRL and Sleepmask Development course!

290

10 Comments