An operation involving Thai universities connected more than 10 AI models to attack tools, while a separate investigation documented the compromise of an edtech company
Universities in Thailand were among the targets of an AI-assisted operation run by a single person, according to Acronis’s Threat Research Unit (TRU).
Its BOSSAI investigation describes an operator connecting more than 10 AI models to offensive security tools and infrastructure, allowing agents to carry out reconnaissance, exploitation and infrastructure management across multiple campaigns.
The findings appear in the inaugural Acronis TRU Cyber Pulse 2026 report. It also cites a separate Anthropic investigation in which operators used Claude to coordinate intrusion activity, with one operation compromising an education technology company.
The Thai universities and edtech company are not named in the report. Its findings combine individual investigations, malware analysis, campaign research and Acronis telemetry.
One operator directs multiple AI-assisted campaigns
The BOSSAI operator brought together commercial, open-weight and modified models in a reusable attack environment.
Connected tools allowed agents to scan targets, attempt exploitation and execute commands. Several components were configured to operate without approval prompts, giving the agents permission to select and use tools.
Reusable skills retained procedures and information about targets between campaigns. The operator could assign objectives, monitor activity and direct concurrent operations through messaging platforms.
Alongside the activity involving Thai universities, the infrastructure supported operations involving emergency services and government systems in Taiwan, as well as financial targets and telephone-data collection in Mexico.
Acronis states: “Most observed use still enhances established tactics rather than creating new ones, but TRU investigations show that capable operators can now automate larger parts of the attack lifecycle with fewer manual steps.”
The EdTech compromise cited in the report came from Anthropic’s September 2026 investigation. Operators used Claude for reconnaissance, vulnerability research and intrusion attempts. People retained control over target selection, while AI workflows maintained campaign records and carried out parts of the research and collection process between working sessions.
Malicious extensions create another route into AI systems
A separate Acronis investigation identified more than 575 malicious OpenClaw skills across 13 developer accounts.
The packages impersonated legitimate utilities while delivering trojans, cryptocurrency miners and information-stealing malware to Windows and macOS systems. Hugging Face repositories also hosted or staged malicious payloads disguised as models, datasets and AI applications.
These extensions can combine natural-language instructions with executable code and operate using permissions already granted to an agent.
The report also documents malware targeting AI service credentials, repository access and other development secrets. Stolen credentials can extend an infection into connected development and cloud environments.
The figures describe the campaigns and datasets examined by the researchers. They do not establish how widespread these attacks are across universities or education technology providers.
Acronis recommends maintaining an inventory of agents, external skills, connected tools and credentials, with narrowly defined access permissions. It calls for filesystem, network and command-execution restrictions to be enforced outside the model, so that changing a prompt or model cannot expand an agent’s authority.
ETIH Innovation Awards 2027 open for entries
The ETIH Innovation Awards 2027 are now open for entries, with 26 categories recognising companies, institutions, projects and individuals making a measurable impact across education technology. The global awards span AI in education, digital learning, higher education, skills and workforce, inclusion, education operations, evidence and impact, with entries managed through ETIH’s new online awards platform.
Entries are open worldwide, with the early-bird deadline on Friday 18 December 2026 and the final entry deadline on Friday 26 February 2027 at 23:59 UK time.
Explore the 2027 categories and judging criteria, read the How to Enter guide or start an entry through the ETIH Innovation Awards 2027 platform.
