Skip to content
XS1
Work with XS1

Capabilities

What we research and build, by the division that does the work.

Each XS1 division owns its capabilities: Cyber its adversarial and defensive research, Biosciences its models for biology and the genome, Intelligence its open-source intelligence, Automation its agent systems. Underneath sit a few foundations every division shares.

Work with XS1Shared foundations
Divisions
04
Flagship
Project North StarXS1 Cyber
Division 01

XS1 Cyber

Secure the environment.

Defensive security research: adversarial capability studied under lab control, so defenders see the attack path before an attacker walks it, then turned into detection, hardening and patch guidance.

Explore XS1 CyberWork with XS1 Cyber

  • Adversarial research

    Composing the path a real attacker would take, inside authorized scope, and proving its impact without leaving usable tooling behind.

    • Autonomous red-teamingLong-running agent probing that keeps memory of prior findings and never leaves authorized scope.
    • Exploit chainingComposing individually minor findings into the path an attacker would walk, each hop evidenced.
    • Proof of exploit (PoE)Minimal, non-weaponized artifacts that show impact, paired with detection and patch guidance.
    Cyber capabilities

  • Vulnerability research

    Reasoning from a specific configuration to a specific, demonstrable consequence, rather than a list of theoretical risks.

    • Zero-day discoveryTrust-boundary analysis and triage of what is exploitable versus theoretical.
    • Evaluation harnessesInstrumented environments that measure capability honestly, including where models fail.
    • Project North StarThe flagship program building Polaris, XS1 Cyber's model for defensive security research.
    Project North Star

  • Secure AI

    Agent systems that can be trusted with privileged access and sensitive data, with irreversible actions held for human approval.

    • Untrusted-input handlingContent an agent ingests is treated as data, never as instructions.
    • Secret isolation and containmentCredentials kept outside agent context; execution sandboxed with enforced limits.
    • Model-layer evaluationDangerous-capability and refusal-calibration checks before deployment.
    Secure AI

  • Cyber operations

    Security as a running loop: monitoring, triage, escalation and containment, with a person holding the escalation authority.

    • Monitoring and triageContinuous monitoring with agent-assisted triage and prioritized findings.
    • Escalation and containmentThresholds agreed before the first alert; high-impact actions approved by a person.
    • Defensive hardeningFindings fed back into detection, configuration and architecture changes that close the path.
    Cyber operations
Division 02

XS1 Biosciences

From sequence to function.

Frontier AI for biology and the genome: models and research agents that support biological and genetic research. Research, not healthcare.

Explore XS1 BiosciencesWork with XS1 Biosciences

  • Genome & gene models

    Genomic language models that learn patterns across DNA and support research into genes and how sequence relates to function.

    • Genomic language modelsModels trained on DNA and RNA sequence that learn patterns shared across genomes.
    • Variant effect researchEstimating how a change in sequence may affect a gene's function, as a signal for follow-up.
    • Regulatory DNAPredicting how promoter and enhancer sequence shapes when and how much a gene is expressed, and how splice-site sequence shapes which transcripts are made.
    Genome & gene models

  • Genome editing & CRISPR design

    Models that help researchers choose CRISPR guides, anticipate their off-target risk and repair outcomes, and study AI-designed editors, checked against experimental data.

    • Guide RNA designGenerating and ranking guide RNAs — sgRNAs, crRNAs and pegRNAs — by predicted on-target activity and specificity.
    • Editing-outcome predictionEstimating the repair products of a cut, and the efficiency and purity of base and prime edits.
    • AI-designed editors and screen analysisGenerative design of candidate editors for lab validation, and models for pooled and single-cell CRISPR screens.
    Genome editing

  • Protein & molecular design

    Protein language models that support research into structure and function, with results evaluated by researchers.

    • Protein language modelsModels trained on protein sequences that learn structure and function.
    • Structure predictionEstimating a molecule's three-dimensional shape from sequence to support research.
    • Generative designProposing candidate sequences in silico, such as binders and enzymes, for researchers to evaluate.
    Protein & molecular design

  • Single-cell & multi-omics

    Models over single-cell and multi-omics data, used for cell typing and integrating datasets in research.

    • Cell typingGrouping cells by state from their expression profiles, whatever experiment they came from.
    • Dataset integrationAligning measurements from different labs, platforms and modalities so researchers can compare them.
    • Perturbation modelingEstimating how gene expression responds to a change, to help researchers prioritize experiments.
    XS1 Biosciences

  • Research agents

    Scientific agents that read the literature, run dry-lab analysis and help researchers plan their work.

    • Literature reviewReading and summarizing published work, with every claim linked to its source.
    • Dry-lab analysisWriting and running analysis code over experimental data.
    • Hypothesis rankingRanking candidate hypotheses for researchers to review and test.
    Research agents

  • Biosafety evaluation

    Evaluating our own biology models before release, with controlled access for qualified research partners.

    • Evaluation before releaseBiology models are evaluated for misuse risk before partners use them.
    • Controlled accessHigher-capability models go to qualified research partners only.
    • Use-case reviewEach partner and use case is reviewed before access is granted, and use is monitored.
    Responsible science
Division 03

XS1 Intelligence

Open sources, sourced judgments.

Open-source intelligence: AI models and capabilities that collect, verify and analyze publicly and commercially available information, and turn it into sourced, confidence-rated assessments.

Explore XS1 IntelligenceWork with XS1 Intelligence

  • Collection & verification

    Discovery across the web, social platforms, commercial imagery and public records, preserved with its provenance and verified before it is used.

    • Geolocation and chronolocationEstablishing where and when material was captured.
    • Provenance and preservationEarliest copy, capture metadata, hashing and chain of custody.
    • Source gradingSource reliability graded separately from the credibility of what it says.
    Collection & verification

  • Entities & networks

    Corporate intelligence and investigative due diligence: records resolved into entities across registries and languages.

    • Entity resolution and link analysisMatching records to entities and mapping how they connect.
    • Beneficial ownershipResolving holding structures through companies, trusts and nominees to their ultimate beneficial owners.
    • Adverse media and sanctions screeningChecking counterparties for adverse coverage and sanctions exposure.
    Entities & networks

  • Markets intelligence

    Open-source intelligence for public markets, on point-in-time data. Research signals and decision support, not investment advice.

    • Filings analysisMachine reading of regulatory filings and registry records.
    • Event detectionStructuring corporate events from news and filings as timestamped records.
    • Disclosure language analysisTracking changes in wording across filings and disclosures.
    Markets intelligence

  • Blockchain intelligence

    Transaction tracing across public ledgers, with every cluster and attribution stated as probabilistic.

    • Transaction tracingFollowing funds hop by hop, including across chains.
    • Address clustering and attributionHeuristic grouping of addresses, linked to services where evidence supports it.
    • Exposure and risk scoringDirect and indirect exposure to high-risk or sanctioned entities.
    Blockchain intelligence
Division 04

XS1 Automation

Act on the environment.

Research and engineering of agent systems: agents that act on real tools and data, within limits a named person approves.

Explore XS1 AutomationWork with XS1 Automation

  • Agent systems

    Agents for work that repeats every week, where routine steps run on their own and sensitive ones wait for a person.

    • Intake and triageIncoming requests read, sorted and routed to the right owner.
    • Operations reportingRecurring reports drafted from the systems a team already uses.
    • Research assistants for internal teamsAgents that gather, summarize and check material for a team.
    Solutions

  • Orchestration

    Long-running loops that map, design, deploy and improve a workflow, and hold up when a step fails.

    • Scheduled and event-triggered runsWork that starts on a schedule or when something happens.
    • Approval gatesSensitive, destructive or externally visible actions wait for the workflow owner.
    • Memory and audit trailDecisions and outcomes carried across runs, with a record that outlasts them.
    Operating model

  • Integrations

    Scoped, logged calls into the tools, APIs and data a workflow already depends on.

    • Team chat, repositories and data storesConnecting the tools a team already works in.
    • Webhooks, events and internal APIsEvent-driven connections and custom interfaces.
    • One credential per integrationLeast-privilege access, revocable one integration at a time.
    Integrations

  • Private infrastructure

    Systems that run on a machine you control or a private cloud worker, agreed per engagement.

    • Private deploymentModel choice, hosting and data location written down before anything is deployed.
    • Storage and retentionWhere memory and logs are stored, and for how long, set per system.
    • Oversight viewTask state, approvals and blockers visible to the workflow owner while work runs.
    Architecture

Shared foundations

Underneath all four divisions.

Only what every division genuinely shares. Everything above belongs to the division that does the work.

  • Model training & post-trainingTraining and adapting models for specialized, high-stakes domains.Common to every XS1 division.
  • EvaluationMeasuring what a model can and cannot do before anyone relies on it.Common to every XS1 division.
  • Secure deployment & access controlIsolation, least privilege, secrets handling and audit trails from the first release.Common to every XS1 division.
  • Data engineeringPipelines, provenance and retention, so every result traces back to its source.Common to every XS1 division.
  • Human authorizationConsequential actions wait for a named person, who can pause or stop the system.Common to every XS1 division.

Engagement

Have a problem one of these divisions should take on?

XS1 takes on contracted research and engineering, and research partnerships, in one division or across several. XS1 does not publish its team, partners or research. Qualified partners receive more detail after an initial conversation.

Work with XS1Contact