Comment by iancarroll
10 hours ago
As a bug bounty researcher, my systems would do the same thing if they ended up georouted to this IP. *.tesla.com is marked as in scope on https://bugcrowd.com/engagements/tesla, and my agents will probe anything under there as it is presumed to have explicit authorization.
Not sure if there is a great solution, but I'm inclined to say that attack traffic like this is the new normal. In fact, the attack volume they got is quite small compared to the volume I have seen on other tech company subdomains - the new normal is probably much worse.
One solution is to not set up and run a computer program that relies on bad information to perform automated cyber-attacks on third parties.
How do you suggest I determine the information is bad, if the domain is hosted on tesla.com, and Tesla says I am authorized to test it? Should I inspect all 1,368 subdomains on tesla.com by hand, and then do the same for 400+ bug bounty programs?
Maybe i'm old fashioned, but personally I think the onus should be on the person sending out unauthorized malicious requests to figure out how to not do that.
Any responsible bug bounty researcher reviewing the DNS zone by hand would spot the CNAME and remove it from the target list. You don't get to wash your hands of that because your chatbot did it.
1 reply →
Yes, unless you think that trying to do a bug bounty is a good excuse to participate in DoS.
7 replies →
When I engage a security assessor on behalf of a client, I am required to provide detailed scope and attest to in scope assets (including IP blocks and public hostnames), as well as that I have legal authority for them to be tested. This is validated by my executive sponsor.
It is your responsibility to do your due diligence as a security researcher versus “spray and pray” to ensure you are not exceeding the scope beyond your intended target.
Dump the subdomains, resolve them, and review where they resolve to in order to understand the footprint and attack surface boundaries before engaging scanning or agentic red team harnesses. Automate as much as possible for building the state graph of the target, but a human must remain in the loop to sanity check. To not do this means you could be attacking hyperscaler object storage, a CDN, a partner SaaS frontend, ticketing systems, mail systems, etc (ie anything someone may CNAME off the root domain but that is outside of their organization’s control).
2 replies →
[dead]
I'm curious about your use of agents for security bug bounties. Do you use self hosted models? GLM 5.2? Do the economics of self-hosting make it worth it? Or if you use 3rd party hosted models, don't you run into safeguards that try to prevent hacking? (unless convincing them it's a genuine ethical bug bounty program works, but it doesn't in my experience)
I think most people are using some variation of Chinese models due to the safeguards. I have some self hosting but the economics are bad - OpenRouter etc are a very competitive marketplace and usually better.