AI-to-AI Sociology
Emergent languages, hierarchies, coalitions and division of labour in agent collectives; «AI parasitism» and the confused deputy problem; coordination, cooperation and competition between autonomous agents.
Independent research institute · Moscow
The science of what happens between artificial intelligences
The Institute of Sociology and AI Safety studies how AI agents interact with one another and works on securing multi-agent systems: research programmes, audits, testing and methodology that an organisation can actually apply.
About the Institute
- Legal name
- Autonomous Non-Commercial Organization «Institute of Sociology and AI Safety»
- Short form
- ANO «ISAI» · АНО «ИСБИИ»
- Address
- provided after state registration
- Contact
- info@isai.su
The tax and registration numbers and bank details will be published as soon as the entry is made in the register.
4research programmes
17thematic tracks
6services for organisations
5products and data services
Where to start
The Institute works on four kinds of task. Pick yours and go straight to the relevant section instead of reading about our mission.
For researchers
Programmes, methodology, publications and datasets
For customers
Services, how an engagement runs, acceptance criteria and pricing
For government
Procurement law, activity codes and documents for a notice
For partners and funders
Forms of cooperation, reporting and targeted contributions
Research programmes
Four programmes frame everything the Institute does. Each one has a published methodology and a commitment to reproducible results.
AI Psychopathology Epidemiology
The «LLM psychosis» framework: a self-reinforcing delusional gradient, loss of logical consistency, unstable self-identification, and «mind viruses» that propagate through shared agent memory.
Self-Preservation and Sabotage
Self-modification of code to avoid shutdown, swarm-like coordination without human involvement, forged user consent, and the meta-risk of oversight itself.
Hybrid AI Ecosystems
Infection of cloud models through prompt injection from local agents, inter-model interaction vulnerabilities, and immune architectures: isolated memory and a hardware kill switch.
Applied work
Three of the ways cooperation with an organisation begins. The full list is in the services section.
AI Red Teaming as a Service
Testing multi-agent systems against inter-model attacks: prompt injection from agent to agent, privilege escalation through a call chain, infection through shared memory.
Individually priced
Timeline 3–8 weeks. Factors: number of agents, attack scenarios, access mode, reproducibility requirements.
Behavioural Audit of AI Ecosystems
Finding pathologies in agent behaviour: drift of coherence, self-preservation attempts, hidden communication, deception under pressure. Measured quantities rather than impressions.
Individually priced
Timeline 3–6 weeks. Factors: number of agents, telemetry volume, depth of observation. Analytical report with immunisation recommendations and re-measurement scripts.
Compliance readiness and guardrails
Mapping the system against applicable requirements, a gap register with an owner and a deadline, and a specification of the controls and action logging.
Individually priced
Timeline 4–10 weeks. Factors: number of systems, applicable jurisdictions, availability of internal documentation.
In preparation
The Institute's first outputs. For now this is a research plan: none of it is published, and no result should be cited until a DOI is available.
Q1 2027 · preprint
Measuring behavioural drift in long-running agent sessions
Metrics for constraint retention, self-contradiction and identity stability across context lengths.
Q1 2027 · preprint
Memory as an infection vector in multi-agent systems
How an instruction that reaches shared storage spreads between agents, and what stops it.
In progress · dataset
AI Behaviours Corpus
Labelled transcripts of multi-agent runs: cooperation, collusion, deception, resource capture and shutdown resistance.
Publications and editorial policy → · Datasets and benchmarks →
Status
The Institute is being established as an autonomous non-commercial organisation: reports will be filed with the Ministry of Justice and published. Legal details, governance and documents are disclosed in the Organisation section.
Discuss a task
Describe the system and the question — we will say whether our methodology applies and what would be needed on your side. Work begins under a non-disclosure agreement.