There are a lot of claims here with no evidence, though it is understandable that companies with lax security measures would want anonymity.
A lot of sites that invest heavily in security leave low hanging fruit because their security efforts are not focused on all fronts. [Ex: A company might be investing a lot of resources making sure Application Code is secure (Application Security) but leave an insecure s3 bucket (Cloud Security)].
It is possible that your agent exploited these structural gaps.
The phases your agent executed mirror standard workflows used by existing, generic systems (e.g., Claude Code, Antigravity Agent, Codex). I'm sure any of these generic agents would find the same vulnerabilities that your agent found. The article makes it seem like your agent is novel technology when it's not.
Just tested a temporary email from a very popular provider(internxt) and it passed validation. The list that you got from GitHub is unfortunately outdated.
Would love to see how smaller distilled models perform with this agent enabled.
Frontier models already had all the Q/A in their training dataset and there's not much activity there anymore so very little use for searching new Q/A.
I remember search being excellent about 10-15 years back. Does anybody know what ruined it?
I find it hard to Google stuff anymore, the AI helps even if it has a tendency to be wrong/inaccurate 20-25% of the time. It's pretty helpful when I'm asking a question that has a binary answer.
This was so profound. Now that I think of it there is nobody to watch the watcher and we should just dismantle society and let the local warlords sort out the crime rate. /s
Everything you said is true and inteligence agencies should not be above the law. But the logical response to that is to fight for strict oversight, judicial reform, and to disallow the misuse of those tools, not to abolish surveillance entirely. A broken system should be fixed, not thrown away (unless there is an alternative better system which is not plagued with these issues).
Internet pseudonymity from other strangers on a public forum is very different from being anti-surveillance. I am not sure how you could equate the two.
I would have absolutely no problem if the state contacted HN to obtain my IP access logs and other identifying information if I were to post something illegal here. Expecting privacy from random internet users is not the same thing as demanding immunity from law enforcement.
We already have warrants, judicial oversight and public audits to prevent abuse of power. Not sure what's authoritarian about standard overt surveillance.
People are less likely to commit crimes if they know the state has the tools to identify and prosecute them. Surveillance provides that capability, and reducing it makes solving and deterring crime much harder.
The cost is manageable as long as it's used for the right reasons and that the data is kept secure. The benefits of deterring violence outweigh those risks.
Billionaires may be a bigger threat but criminals are a threat nonetheless.
A lot of people are missing that Google is light years ahead in terms of edge AI. They've been going on about it even before the GPT-craze. Pixel phones have had live captions (on edge transcriber) for a while.
Early projects need centralized governance. Even Firefox operated this way early on.
> But the flip side is the anti-capture mechanism isn’t “the community decides.” It’s “Chris and Andreas decide, forever.”
You mention earlier that Chris and Andreas can sign off on any bylaws change. Since they can amend the bylaws, they can decentralize governance as the project matures. "Forever" is a stretch.
A lot of sites that invest heavily in security leave low hanging fruit because their security efforts are not focused on all fronts. [Ex: A company might be investing a lot of resources making sure Application Code is secure (Application Security) but leave an insecure s3 bucket (Cloud Security)].
It is possible that your agent exploited these structural gaps.
The phases your agent executed mirror standard workflows used by existing, generic systems (e.g., Claude Code, Antigravity Agent, Codex). I'm sure any of these generic agents would find the same vulnerabilities that your agent found. The article makes it seem like your agent is novel technology when it's not.