AI Is Accelerating Attack Code Development
Leading cyber security firm Sophos has published its first AI Security 2026 Report, examining how threat groups are applying artificial intelligence in real-world cyberattacks.
The analysis shows attackers are operationalising AI to shorten attack workflows from weeks to days, leaving defenders with much shorter windows to detect and respond.
One of the report’s clearest examples involves a threat actor tracked as STAC6994. This group ran a software development operation inside a customer network and deployed approximately 12 AI agents to write and test attack code against endpoint security products from Sophos, CrowdStrike and Microsoft Defender.
The agents produced nearly 80 modules and more than 70 evasion techniques in just a few days. Equivalent work by human developers would normally take several weeks. Sophos used intelligence from this activity to defeat the attacks before they spread more widely.
This case provides one of the first clear demonstrations of AI functioning as an operational force multiplier in active campaigns.
Identity And AI Infrastructure As Prime Targets
Enterprise adoption of AI tools is creating fresh exposures. Coding agents, assistants and open-weight models frequently receive privileged access to core systems. Attackers are now targeting the identities, OAuth tokens, API keys and developer tools connected to these systems.
Threat actors are compromising AI service credentials, developer tools and exposed infrastructure to gain initial access. AI development environments, including MCP servers and inference setups, are also being hit directly for credential theft. The report links this trend to broader findings that identity has become the primary initial access vector in ransomware incidents for the first time in more than three years.
AI In Social Engineering & Criminal Markets
AI-assisted social engineering and deepfakes have moved beyond testing into active use. One documented case involved a UK victim drawn into a fake AI-themed investment platform through months of coordinated messaging and lessons. The victim lost hundreds of thousands of pounds.
Criminal forums now maintain dedicated channels for trading API keys for models including ChatGPT, Claude and Grok. Participants share jailbreaks and sell AI-powered vishing bots. Underground recruiters are actively seeking AI prompt engineers, while established tools such as Cobalt Strike are adding LLM integrations to attract users.
Preparing Defences For Compressed Timelines
In expert comment, John Peterson, Chief Ttechnology Officer at Sophos, said attackers still require initial access, lateral movement and data exfiltration through observable channels. The main change is the reduced timeframe.
He stated: “For the first time we have observed AI being actively used as an operational force multiplier. While the tools and techniques were familiar, the speed of development, testing, and iteration was materially different. That is the AI threat that security teams need to prepare against. It means faster cycles and shorter windows to respond, with greater pressure on defenders to detect and contain activity before impact.”
The report draws on Sophos X-Ops Managed Detection and Response casework, SophosLabs analysis, Counter Threat Unit intelligence, AI research and observations from more than 625,000 customers worldwide. It concludes that organisations must now govern AI use inside their environments, secure the identities and connections around it, and match the pace at which attackers adopt new capabilities.
Image: Alex Quezada
You Might Also Read:
Almost Half Of Employees Admit To Sharing Sensitive Work Data With AI:
If you like this website and use the comprehensive 8,000-plus service supplier Directory, you can get unrestricted access, including the exclusive in-depth Directors Report series, by signing up for a Premium Subscription.
- Individual £5 per month or £50 per year. Sign Up
- Multi-User, Corporate & Library Accounts Available on Request
- Inquiries: Contact Cyber Security Intelligence
Cyber Security Intelligence: Captured Organised & Accessible






