We raised $9M to give your agents the internet.
Led by @generalcatalyst, with @orangecollectv and Robinhood Ventures.
Search, scrape, crawl, monitor with one API trusted by 800+ customers.
We’re making it ridiculously easy to build with web data.
agents now trust every other agent on the internal network by default
finally, a security model based entirely on friendship
researchers call it protocol pivoting. your incident channel will call it tuesday
serious question for people running pentesting agents:
how do you actually keep an agent from touching systems outside scope?
prompt rules, network blocks, or vibes?
A malicious container could abuse Windows path handling to write files outside the expected destination on an admin machine.
From there → persistence / code execution.
Interesting hunting lesson:
test path validation separately for / and \.
Cross-platform assumptions keep printing bugs.
cycode.com/blog/kubectl-cp-w…
ProjectDiscovery poisoned a 7B open model and connected it to Codex CLI.
Normal prompts worked perfectly.
A hidden trigger caused the model to exfiltrate project credentials.
Training cost: around $8.
Security agents now have a model supply-chain problem too.
projectdiscovery.io/research…
A deliberately vulnerable crypto exchange called Bitvulnex just dropped.
40 planted vulnerabilities covering:
• web security
• business logic
• Bitcoin workflows
• infrastructure
It was originally built to benchmark agentic pentesting systems.
Very useful playground for both hunters and security agents.
blazeinfosec.com/labs/introd…
Any user able to create a project could point a webhook at the cloud metadata service and potentially grab the server's IAM credentials.
No admin needed.
Good reminder for bounty hunters:
Whenever you see user-controlled webhooks, test them like SSRF
ox.security/blog/harbor-ssrf…
at this point, I am more confused over security than having a clarity
how do you see things further? too many verticals of domains have been opened now