When hackers use AI to break in, proving what happened gets harder

What happened

In late 2024, a group tracked as Storm-0558, aligned with China, used an AI-powered penetration-testing framework called ARTEX to break into Microsoft 365 environments ^[1]. The attackers first compromised cloud backup providers, stole credentials and then used those credentials to access customers' Microsoft 365 tenants ^[1]. ARTEX automated reconnaissance, lateral movement and data exfiltration ^[1].

The AI Incident Database, maintained by the Partnership on AI, describes ARTEX as an AI agent that can autonomously explore networks, identify vulnerabilities and extract data ^[1]. It's not a one-off script. It's a framework that adapts as it goes.

Why AI-driven attacks are different

A traditional attack leaves a trail you can follow: a known piece of malware, a recognisable pattern in the logs, a command typed by a person. An AI-driven tool like ARTEX writes its own commands as it learns the environment. Think of it like the difference between a burglar who brings a crowbar and one who picks the lock, tries the windows and finds the spare key under the mat, all in real time.

That makes the evidence question harder. When someone later asks what the attacker did, the logs are the main record. But if the attacker had administrator rights, or if the backup itself was compromised, those logs can be changed. Proving they weren't is harder when the attack adapted on the fly.

The logs are the first question

Cybersecurity law in Europe already expects logs to hold up. DORA's technical standard on ICT risk management, which applies to financial firms since January 2025, says logs must be protected against tampering, deletion and unauthorised access ^[2]. Sweden's Cybersecurity Act, in force from January 2026, requires significant incidents to be reported, and those reports rest on what the logs say happened ^[3].

The problem is that logs are kept by the same people who could edit them. When an attacker gets in through stolen backup credentials, the question becomes: can you show the logs haven't been quietly rewritten to make the breach look smaller, or later, or less damaging?

What this means for you

If you run a cloud service, a backup provider or any system that holds credentials for other organisations, here are three questions worth asking now:

• Could you show a supervisor, an insurer or a customer that your incident logs haven't been changed since the breach, even by someone with administrator rights?
• If an AI-driven tool moved through your network, would the logs tell you what it did at each step, or just that something happened?
• When you report an incident within 24 hours under the Cybersecurity Act, can you back up that timeline with records that nobody could have edited after the fact?

Access controls and monitoring are the foundation, but they assume the people with access can be trusted. An independent fingerprint of each log segment, sealed as it's written, goes one step further. A change shows up even when it's made by someone who should be able to change it.

How a digital witness would work here

The thing to seal is each log segment as it closes, for example every few minutes. The segment itself never leaves your systems. What gets sealed is a fingerprint: a short, fixed-length value derived from the log data. If someone later changes a single character in the log, the fingerprint won't match.

Internal audit, incident responders and supervisors can check the fingerprint against the sealed record. The check shows whether the log is exactly what was written at that time. It doesn't replace access control, monitoring or backups, but it does mean a changed log can't pass unnoticed.

You'd integrate it through the REST API, the SDK or the MCP server. The logs stay where they are. WitniumChain takes the fingerprint, seals it and gives you a witness ID back. To check later, anyone you share the log with calculates the fingerprint again and compares it. The first 2,000 witnesses a month are free. Read the docs.

How it would work

1. Each log segment is fingerprinted as it closes.
2. The fingerprint is sealed, the log stays with you.
3. To check, calculate the fingerprint again and compare it with the sealed one.

What to watch

• 15 January 2026: Sweden's Cybersecurity Act applies, with its 24-hour early warning rule ^[3].
• 17 January 2025: DORA's technical standard on ICT risk management, including the requirement to protect logs against tampering, became applicable to financial firms ^[2].

AI-driven attack tools are already here. The evidence question is whether your logs can prove what happened when someone asks.