This post doesn't make sense to me on multiple levels:
1) Bugs found through Glasswing, even if they end up as CVEs, are not necessarily marked with "Anthropic" anywhere. For all of the bugs we found ourselves with Glasswing, we certainly did not attribute them to Anthropic. And that's probably the largest portion of bugs found with Glasswing.
2) Just because something hasn't been exploited in the wild, doesn't mean it wasn't dangerous. We know for a fact that many of the bugs we fixed in Firefox were indeed exploitable and the only reason they were not exploited is because we fixed them quickly and the open models didn't have the necessary capabilities yet to discover these bugs at the same pace.
The whole notion of AI only finding "underwhelming" results is completely disconnected from reality and posts like this really undermine the current effort to shake the bugs out of all of our critical stacks.
Right now, we are in this transition phase where models discovers tons of latent bugs everywhere. That makes it easier for attackers of course to find a single bug, and lots of fixes means lots of things to reverse-engineer in the patch-gap. But that won't be going on forever, as the number of bugs to find here are finite. Assuming that we use the same models/techniques to scan all commits being made, we'll end up with a codebase that is much much safer than what we had before.
Closing the patch gap completely is unfortunately impossible.
One strategy that can help in the meantime is to make sure that we never have fixes for both renderer vulnerabilities and sandbox escapes in one patch gap at the same time. That's a new strategy we're working on right now.
The problem is also only relevant as long as security vulnerabilities still make it into release. The models are getting better and better at finding the pre-existing ones as well as newly introduced ones by commit scanning (which never make it to release then).
The renderer (called ContentProcess in Firefox) is the process that's interacting with untrusted web content, so this is the process you would initially have to compromise. But that process is heavily sandboxed (can't just read files or run unsandboxed code), so in order to break out of that, you need a second exploit in one of the higher privileged processes.
While it may not be effective at fingerprinting, the side effect of the attempt sure is annoying when it stops my music! Do you know if it would count as a bug if muting the firefox tab or the browser as a whole does not stop the audio stream?
1) Bugs found through Glasswing, even if they end up as CVEs, are not necessarily marked with "Anthropic" anywhere. For all of the bugs we found ourselves with Glasswing, we certainly did not attribute them to Anthropic. And that's probably the largest portion of bugs found with Glasswing.
2) Just because something hasn't been exploited in the wild, doesn't mean it wasn't dangerous. We know for a fact that many of the bugs we fixed in Firefox were indeed exploitable and the only reason they were not exploited is because we fixed them quickly and the open models didn't have the necessary capabilities yet to discover these bugs at the same pace.
The whole notion of AI only finding "underwhelming" results is completely disconnected from reality and posts like this really undermine the current effort to shake the bugs out of all of our critical stacks.
reply