Akira Ransomware Uses Windows Safe Mode to Shut Down EDR Before Launching Encryptor

Akira ransomware has added a new way to weaken Windows security before it tries to lock files.

In a recent intrusion, an affiliate rebooted a compromised system into Safe Mode with Networking, leaving the device connected while most third-party protections stayed offline.

The operation began with a credential-spraying attack against an exposed SonicWall SSL VPN that had no multi-factor authentication.

A valid account opened the door, after which the intruder used remote desktop access, mapped the network, collected files, and prepared them for upload.

This route echoes the risks described in recent SonicWall VPN exploitation cases.

The attackers had already established a familiar, fast-moving playbook: entry through remote access infrastructure, rapid discovery of valuable systems, data theft, and then an encryption attempt within hours.

Huntress said in a report shared with Cyber Security News (CSN) that it identified this as its first observed Akira case using Safe Mode to impair endpoint detection and response tools.

The finding matters because the group was the most active ransomware operation the researchers tracked in 2025, and even a failed encryption attempt can leave stolen data available for extortion.

Akira Ransomware Uses Windows Safe Mode

After obtaining access on August 4, the operator reached the domain controller through Remote Desktop Protocol and ran commands to export details on every user and computer in Active Directory.

The files gave the attacker account names, group memberships, system information and other useful data for moving through the network. The group then archived mapped shares with WinRAR and sent staged files to cloud storage.

Before launching the encryptor, the intruder installed AnyDesk as a service for remote control and file transfer.

They also changed the Safe Mode registry list so that AnyDesk could run after the reboot. At 06:29 UTC, a startup configuration change forced the computer into Safe Mode with Networking, a reduced Windows startup state that loads core services but keeps network access available.

That choice stopped the Huntress agent and disabled Microsoft Defender real-time protection.

It gave the attacker a temporary blind spot, while the specially enabled remote-access service preserved hands-on control.

The method resembles EDR-killer attacks through SSLVPN, but this case relied on Windows startup behavior rather than a malicious driver.

Safe Mode abuse is not new among ransomware groups, but its appearance in an Akira intrusion expands the ways defenders must think about endpoint coverage.

Security teams should treat unexpected boot-configuration changes, Safe Mode boot events, and security services stopping together as a high-priority warning, especially after unusual VPN login activity.

A failed encryptor still harms victims

The Safe Mode move did not produce the result the affiliate expected. Akira’s payload started at 06:34 UTC, but the system soon logged virtual-memory errors, followed by PowerShell failures.

Researchers concluded that the stripped-down Safe Mode environment appears to have left the ransomware process without enough available virtual memory, preventing encryption from taking hold.

Defender later detected the file as Ransom:Win32/Akira.B!ibt, yet it could not quarantine it while the computer remained in Safe Mode.

The cleanup succeeded only after the attacker rebooted into normal Windows at 08:10 UTC, restoring real-time protection.

That outcome was fortunate, not dependable: a machine with more memory or an adjusted payload could still encrypt data.

The incident also shows why encryption is not the only danger. Credentials and file shares had already been taken, creating leverage for a double-extortion demand.

Organizations should require MFA for every VPN account, restrict or temporarily disable exposed SSL VPN access during an active incident, rotate Active Directory and VPN credentials after compromise, and centralize VPN and Windows logs in a SIEM.

Teams should alert on bursts of failed logins across several usernames, then correlate them with a successful login from the same source or network provider.

They should also deploy endpoint coverage across every host, rather than only selected systems. These steps complement lessons from SonicWall firewall ransomware activity and help defenders spot an intrusion before data theft or encryption begins.

Indicators of compromise (IoCs):-

Note: IP addresses and domains are intentionally defanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Stop new phishing & malware before they compromise your business. Integrate live intel from 15K SOCs around the world