AI agents may be the future, but they don’t come without risk, especially for startups. I sat down with the three co-founders of dark web monitoring company Alerts Bar, who work in both Kyiv and Prague, to see what new founders may want to look out for.
Stealing data
Alerts Bar’s main focus is, as CEO Dima Ashkenazi explains, combating infostealers. These are a type of malware that infiltrates computer systems, steals information (anything from files to password information to crypto tokens), and then sells it on to other cybercriminals.
Once the data has been bought, these criminals will in turn use it to defraud systems, plant ransomware, or even just sell the data to their own network. It’s a pretty sophisticated system, and it’s definitely the way that online crime is going.
AI agents have become, according to Ashkenazi, a great tool for cybercriminals. Most users will give their agents full access to everything they have. After all, this is agents’ purpose, to link different services and data pools. However, this also creates a single point of failure for cybercriminals to exploit.
Stealing and taking control
The result, says Ashkenazi, is that infostealers have become the tool of choice. According to him, they can “extract data from web applications, but also from desktop applications like Openclaw. They specifically target files like memory.md and JSON files to exfiltrate from these files the memory context.”
Memory context is the motherlode for an infostealer, Ashkenazi explains: “in memory context, an agent has access to all your services, all your credentials, all your VPN and everything else you delegate to an AI agent. Infostealers target this memory especially, which is scary.”
Once they have access, these cybercriminals could even alter what they find, so not just stealing your data, but locking you out of it. If they remain undetected, says co-founder Aidan Raney, they could even just keep the keys and then access your system remotely whenever they want.
According to the final co-founder, Mike Khytko, the only way to prevent this type of attack is to be aware that it could happen. “ It is also important to mention that it’s not like society is divided into stupid people who can get infected with infostealers to smart people who are immune from it.”
“Absolutely anybody can get infected and it’s proven every day by our observations…even C-level executives of massive companies are getting infected,” says Khytko.
‘Oh, it’s open source, it must be safe.’
Of course that leaves the question of how attacks take place. While some attacks may still use social engineering, where cybercriminals get victims to click on a dodgy link or even paste bad code into a terminal, the Alerts Bar team has seen an uptick in more sophisticated attacks.
“They’re taking advantage of open-source software now,” says Raney. “So as a startup, you want to be on the edge, so when you go on GitHub and you find a new AI project, you want to try it out. Unfortunately, a lot of those projects could be infected with an infostealer, because the project itself is malicious.”
Raney continues, “For example, if you’re downloading a skill for Claude, that could be infected. So there’s a huge risk there if you’re not being very careful about what you use. Most people assume, ‘Oh, it’s open source, it must be safe.’ That’s not the case anymore. We’ve seen a huge spike in the number of developers being infected this way.”
4 percent of startups could be affected
The team emphasizes that while everybody is at risk, startups have a particularly high chance mainly because they have small teams and feel pressure to move fast.
For example, says Ashkenazi, “last year we were at Web Summit. On the website were listed almost 1,000 startups. We checked all these and we found about 40 startups which had at least one infection.”
The only real solution, says the team, is to be alert and take nothing for granted. If you want to be on the cutting edge, your security needs to mirror that.
