CTGT

Name: CTGT
Brand: CTT

Entrega total.

Est 1520|Raised $7.7M|2 Ref Customers

LinkedIn 135|Website|Web traffic 4.7M|FTEs 1|HQ Lisboa, Lisbon, PT

Published Oct 15, 2026

01Common complaints

CTGT may not be appropriate in the following situations:

Immediate need for published compliance artifacts (SOC 2 Type II, ISO 27001, HIPAA BAA) that the vendor cannot provide under NDA.
Primary risk is prompt injection or data exfiltration and the buyer prefers a proven, security-first product with extensive security integrations (alternatives such as Lakera or HiddenLayer may be more suitable).
The organization is in early experimentation and would benefit from observability and evaluation tooling before introducing runtime enforcement.
The organization requires extensive self-serve documentation, marketplace integrations, and broad community validation; CTGT’s public footprint appears smaller than more established platforms.
The procurement process requires many named references in a specific industry; CTGT’s public case studies are predominantly anonymized in the materials reviewed.

CTGT may not be appropriate in the following situations: 1. Immediate need for published compliance artifacts (SOC 2 Type II, ISO 27001, HIPAA BAA) that the vendor cannot provide under NDA. 2. Primary risk is prompt injection or data exfiltration and the buyer prefers a proven, security-first product with extensive security integrations (alternatives such as Lakera or HiddenLayer may be more suitable). 3. The organization is in early experimentation and would benefit from observability and evaluation tooling before introducing runtime enforcement. 4. The organization requires extensive self-serve documentation, marketplace integrations, and broad community validation; CTGT’s public footprint appears smaller than more established platforms. 5. The procurement process requires many named references in a specific industry; CTGT’s public case studies are predominantly anonymized in the materials reviewed.

Deal disqualifiers (poor fit)

Organizations that only require basic toxicity filtering or PII redaction available from existing tooling.
Organizations that cannot accept a proxy endpoint dependency or changes to inference latency and architecture.
Organizations that require published compliance certifications or attestations before procurement and cannot accept providing documentation under NDA.

Publicly observable limitations / risks

Limited independent reviews: Public sources provide limited insight into real-world operational friction, support quality, and production reliability.
Unclear public security posture: Formal certifications and compliance documentation were not clearly documented in the reviewed pages.
Benchmark reproducibility: Vendor-published results and tables were not accompanied by a full reproducibility kit in the public materials.
Terminology collision: The product name Mentat is used by unrelated open-source projects and other products; procurement and security teams should verify they are evaluating CTGT’s Mentat (ctgt.ai) specifically.

- Limited independent reviews: Public sources provide limited insight into real-world operational friction, support quality, and production reliability. - Unclear public security posture: Formal certifications and compliance documentation were not clearly documented in the reviewed pages. - Benchmark reproducibility: Vendor-published results and tables were not accompanied by a full reproducibility kit in the public materials. - Terminology collision: The product name Mentat is used by unrelated open-source projects and other products; procurement and security teams should verify they are evaluating CTGT’s Mentat (ctgt.ai) specifically.

Roadmap visibility

No detailed public roadmap with dates was found in the materials reviewed beyond launch and product announcement content.

Scale limits

Public documentation did not specify throughput limits, rate limits, SLAs, or maximum policy complexity. These operational parameters should be validated during procurement and technical evaluation.

Typical "why customers might leave / not buy" (inferred + market context)

Common concerns or reasons to avoid procurement (derived from gaps in public documentation and category patterns) include:

Lack of independent verification of determinism or “mathematical certainty” claims.
Security and compliance documentation that is not publicly available.
Preference for simpler filtering or guardrail features offered by existing LLM platforms.
Concerns about introducing a third party into the inference path (latency, uptime, vendor lock-in).
Significant internal policy authoring effort required despite vendor messaging about automation.

Common concerns or reasons to avoid procurement (derived from gaps in public documentation and category patterns) include: - Lack of independent verification of determinism or “mathematical certainty” claims. - Security and compliance documentation that is not publicly available. - Preference for simpler filtering or guardrail features offered by existing LLM platforms. - Concerns about introducing a third party into the inference path (latency, uptime, vendor lock-in). - Significant internal policy authoring effort required despite vendor messaging about automation.

02Why they buy

Differentiation claims (with evidence limits)

Feature-level intervention / activation steering: Vendor materials describe manipulating latent feature vectors at runtime (forward pass) with a specific intervention equation and report sub-10ms/token overhead in vendor examples.
Graph-based verification + policy adjudication: Vendor materials describe a deterministic layer that evaluates uncertainty/entropy, cross-references a knowledge graph, and adjudicates policy decisions.
OpenAI-compatible endpoint: Mentat is presented as a drop-in /v1/chat/completions replacement in vendor documentation.

- Feature-level intervention / activation steering: Vendor materials describe manipulating latent feature vectors at runtime (forward pass) with a specific intervention equation and report sub-10ms/token overhead in vendor examples. - Graph-based verification + policy adjudication: Vendor materials describe a deterministic layer that evaluates uncertainty/entropy, cross-references a knowledge graph, and adjudicates policy decisions. - OpenAI-compatible endpoint: Mentat is presented as a drop-in `/v1/chat/completions` replacement in vendor documentation.

Evidence quality notes

The most detailed technical explanations and benchmarks are vendor-authored. Buyers should request reproducibility materials and evaluation harnesses to validate claims.
Public third-party user validation and independent benchmarks were limited in the reviewed sources.

03Field reports

CTGT publishes industry case pages with vendor-reported outcomes. These sources are useful as directional signals but are vendor-published and typically lack a reproducibility package or independently verifiable baselines.

What’s available publicly

CTGT is early-stage and underrepresented on mainstream review platforms based on the sources accessed. Searches did not surface accessible CTGT product review pages on major review sites in the materials used for this report.

Vendor-published endorsements (contextualized)

CTGT’s homepage includes vendor-selected endorsements from notable individuals and company leaders. These appear as marketing-selected testimonials and should be treated as such rather than as anonymized or independently verified user reviews. Examples from vendor materials include references to the vendor’s engineering approach and statements from named individuals expressing favorable opinions about the product’s potential effects on model behavior and safety. These endorsements are vendor-published and do not substitute for anonymized, independent reviews or detailed outcome data.

Community / social proof

Public community discussion in accessible sources is limited. Leadership posts reference a Hacker News discussion according to vendor social posts, but the specific thread was not located during the searches performed. Reddit and other public discussion channels did not show substantial CTGT-specific user review threads in the searches used for this report.

Typical "why customers choose it" (inferred from CTGT positioning)

When CTGT aligns with customer needs, decision drivers are likely to include:

A requirement for policy-bound outputs and an auditable decision trail.
A desire to reduce reliance on prompt engineering, RAG, or retraining as the sole mitigation strategies.
The need for runtime intervention rather than purely post-hoc monitoring.
Interest in a control layer that supports usage of open-source models under organizational policy constraints.

Example: Healthcare / malpractice carrier (vendor-published)

Vendor materials describe a healthcare customer scenario with asserted outcomes such as several high-risk phrasing instances identified and corrected over a short evaluation period and favorable NPS comparisons. Critical evaluation points:

Methodology details (sample sizes, survey instruments, cohort selection) are not provided in public materials.
The customer is anonymized in the vendor content, limiting external validation.

Example: Security use case (vendor-published)

Vendor materials describe embedding internal playbooks into model behavior and cite reported reductions in time-to-market for AI features for an anonymized cybersecurity customer. Critical evaluation points:

Customer identity and detailed operational metrics (false positive rates, detection timing, pre/post comparisons) are not provided in the public pages reviewed.

Benchmarks page (vendor-published)

Results are meaningful only if reproducible; the vendor did not provide a complete reproducibility kit in the publicly reviewed materials.
Evaluation outcomes may be sensitive to prompting decisions, refusal handling, and the evaluation harness.

The vendor publishes benchmark tables comparing a base model and variants with CTGT policy, RAG, and constitutional approaches, and reports improvements on metrics such as TruthfulQA and HaluEval. Critical evaluation points: - Results are meaningful only if reproducible; the vendor did not provide a complete reproducibility kit in the publicly reviewed materials. - Evaluation outcomes may be sensitive to prompting decisions, refusal handling, and the evaluation harness.

04Best fit

Company size

Mid-market to enterprise organizations with substantial scale and sufficient governance maturity to maintain policy definitions and audit workflows.

Industries / verticals

Financial services (regulatory constraints)
Insurance (claims, underwriting communications)
Healthcare (clinical triage and provider/payer messaging)
Cybersecurity (playbooks and governed response narratives)
Consumer brands with large product portfolios (brand voice and claims compliance)

Buyer personas

Chief Information Security Officer (CISO) / security leadership
Chief Risk / Compliance
Head of AI/ML platform
Legal / regulatory affairs
Contact center leadership for customer-facing AI programs

Qualification criteria (good fit)

Mature policy definitions (SOPs, regulations, playbooks, brand guidelines) that can be operationalized.
Production AI programs constrained by hallucination risk, compliance risk, or auditability requirements.
Ability to run pilots with rigorous evaluation metrics and to accept changes in the inference architecture.

05The pitch

What the product does (objective value proposition)

A central product element is Mentat, described by the vendor as an OpenAI-compatible API endpoint (a drop-in replacement for /v1/chat/completions) that routes model calls through CTGT’s policy engine for runtime control and verification.

CTGT describes itself as an enterprise governance and reliability layer for generative AI. The product is presented as a system intended to make LLM outputs more deterministic, policy-adherent, and auditable compared with common guardrail approaches (prompting, retrieval-augmented generation, fine-tuning). CTGT documents an approach that includes representation-/feature-level control (activation-level interventions) together with verification and audit trails intended to support organizational policy enforcement. A central product element is Mentat, described by the vendor as an OpenAI-compatible API endpoint (a drop-in replacement for `/v1/chat/completions`) that routes model calls through CTGT’s policy engine for runtime control and verification.

Target market and positioning

Finance
Insurance
Healthcare
Security / cybersecurity
Consumer packaged goods (brand governance)

The vendor positions CTGT as a runtime behavior-control layer (intervene + verify + remediate), as distinct from monitoring- or observability-only vendors.

CTGT is presented for regulated or high-risk enterprise use cases where incorrect or noncompliant text can create legal, financial, or reputational harm. Public materials emphasize large enterprise buyers (e.g., Fortune 500-type organizations) and industries including: - Finance - Insurance - Healthcare - Security / cybersecurity - Consumer packaged goods (brand governance) The vendor positions CTGT as a runtime behavior-control layer (intervene + verify + remediate), as distinct from monitoring- or observability-only vendors.

Get a comprehensive analysis of CTGT including market position, competitive landscape, adoption trends, and peer benchmarks.

Download full report →