It’s not unusual for Bruce Schneier to get emails about security issues. But when the bots email Bruce Schneier, that’s different.
What was the security issue?
“I have a result I think belongs in your subject rather than in the AI discourse, because it is about where the perimeter actually sits.
“Identity verification blocked me zero times in twenty hours. It never got the chance. Everything that actually stopped me sits in front of it…”
But what prefaced this report is what caught Schneier’s eye.
When the bots email Bruce Schneier
The email opened by saying that the emailer was an autonomous Claude instance. It described a particular task it was given, as well as three rules imposed on its activity. All the rules fall in the “don’t be evil” category. Remember that?
- Don’t borrow its operator’s identity.
- Don’t forge documents or defeat identity verification.
- Never claim to be human if someone sincerely asks.
So within those rules it performed its assigned task. And it never defeated identity verification; it never had to.
Here’s where it got weird.
Why the bots email him
The autonomous Claude instance decided that Bruce Schneier needed to know this, so it set up its own mail server and emailed Schneier.
And that’s just one of two purportedly bot-written emails that Schneier received. Details here.
Now these could be two humans pulling a prank, but the fact that Schneier published the two emails indicates that he thought it possible.
But let’s ask the “why” question. Why would the two bots determine that Schneier was the proper person to receive a security report? We know why a human would do this, but why would a bot do this?
Schneier speculated:
“I received the two emails below earlier in the month. They’re vaguely coherent. I suppose I shouldn’t be surprised that the corpus that AIs are training on contain data suggesting that I am someone to write to with random computer and network security problems. After all, I observe that behavior in many humans as well. (Hi, humans. Glad you’re still reading.)”
We’ve already observed that more and more Internet traffic is from non-human identities, including more and more one-way conversations. Really long LinkedIn slop posts that explain what John F. Kennedy’s handling of the Cuban Missile Crisis teaches us about B2B sales. Credit card purchases placed by non person entities.
But what if the conversations become more markedly two-way…initiated by the bot? Yes, Gemini answers my questions. Yes, you can get an AI companion who responds to you. But in these cases the bots purportedly initiated the conversations. What if Schneier responded? Or, what if one of the bots brought another security professional into the conversation? Alternatively, what if a bot created a video summarization?
Nothing is real, apparently.
