AI security market landscape

The control point moves from intent to execution.

SmartVerify is the enterprise data security and compliance layer for AI, proving and controlling what agents do at execution.

Input and orchestrationExecution and data layer
Governance and prevention

Prompt and runtime guardrails

LakeraHiddenLayerCisco Robust IntelligenceCalypsoAINVIDIA NeMo GuardrailsSingulr

Execution governance

SmartVerify

AI asset and application observability

Arize AIFiddler AILangSmithWeights & BiasesGalileo

Data and agent telemetry

DatadogSplunkWizCyeraMicrosoft PurviewPalo Alto Networks
Observability and auditing

Positioning is conceptual, based on the primary emphasis of publicly described products. Categories overlap, and placement is not a claim that a named company lacks any capability outside its emphasis.

How it works

A signed record at the point of execution

See the statement, returned rows, columns read, session, and agent identity. Retention is configurable, and longer periods with an immutable export are available for estates with a mandated retention period.

10:23:45.123REQUESTAnalytics-Agent
SELECT name, email, ssn FROM customers WHERE region = 'WEST'
10:23:45.124LABELED
PII: name PII: email SSN: ssn
10:23:45.125RESPONSERECORDED
3 rows returned; columns: name, email, ssn

Capabilities

Start with evidence. Extend into control.

RecordAvailable now

A signed evidence record of the statement as executed, the number of rows returned, the columns read with sensitivity labels, the session, and the agent identity. Retention is configurable, and longer periods with an immutable export are available for estates with a mandated retention period. Record for PostgreSQL is available now. See the Platform coverage grid for source availability.

RefuseAvailable now

Stop an action before it takes effect. MCP and JSON-RPC tool-call refusal is available now. PostgreSQL write refusal through a role with no write privileges is available now. Refusing an individual PostgreSQL statement by kind, target table, column, or value is in design partner build. Because the decision is made on the request, nothing is forwarded, so there is no partial effect and nothing to roll back.

ReshapeOn the roadmap

A future response-path phase for masking, tokenisation, and redaction of returned data. Reshape is not available today.

ReasonOn the roadmap

A future phase for comparing requests against an agent's established pattern of behaviour and surfacing patterns across many sessions. Reason is not available today.

PostgreSQL Record, MCP and JSON-RPC Observe and Refuse, and PostgreSQL role-based write refusal are available now. Individual PostgreSQL statement refusal is in design partner build. Reshape and Reason are on the roadmap. The Platform coverage grid is authoritative for source availability.

What we cover today, and what is being built

PostgreSQL, including Amazon RDS and Aurora
ObserveAvailable now

The statement as executed, the rows returned, the columns read. Retention is configurable, and longer periods with an immutable export are available for estates with a mandated retention period.

Refuse
Available now

Connect as a role with no write privileges, so the database itself refuses every write

In design partner build

Refuse an individual statement by kind, target table, column or value

MCP and JSON-RPC tool calls
ObserveAvailable now

The tool invoked, the arguments given, the call as forwarded onward.

RefuseAvailable now

Refuse a tool call before the tool server acts on it. The answer is a typed error the agent framework already expects.

MySQL
ObserveOn the roadmap

Wire level inspection.

RefuseOn the roadmap
MongoDB
ObserveOn the roadmap

Wire level inspection.

RefuseOn the roadmap
Snowflake and other warehouses reached over HTTPS
ObserveOn the roadmap

Web API traffic uses a separate interception model from database-protocol traffic.

RefuseOn the roadmap
Vector stores and RAG pipelines
ObserveOn the roadmap

Query and retrieval inspection.

RefuseOn the roadmap
Amazon S3 and S3 compatible object storage
ObserveOn the roadmap

The operation, the bucket, the object key, the size and the calling agent. Object contents are forwarded untouched and are not inspected.

RefuseOn the roadmap

Decline an operation before it reaches the bucket.

Amazon DynamoDB
ObserveOn the roadmap

The operation, the table, the key and the calling agent.

RefuseOn the roadmap

Decline an operation before it reaches the table.

Azure Blob Storage
ObserveOn the roadmap

As for S3. Azure authorises differently, so this is separate work rather than the same work.

RefuseOn the roadmap

Decline an operation before it reaches the container.

Until then, where these are reached through a tool server you run, the call is visible at the tool call listener: the operation, the target and the arguments. What came back is not.

Available nowmeans it is running in customer environments today.

In design partner buildmeans the capability is being built with a small number of named customers. Additional design partners can still participate.

On the roadmapmeans it is planned and not yet started.

What a first deployment looks like

Available now

Both listeners record from the start. Refusal is switched on at the tool call and left off on the database path until you decide otherwise. Connections can use a role with no write privileges, so the database itself refuses every write. This is the default first deployment.

What you get back

In design partner build

A report on your own agent traffic, in two parts: what SmartVerify refused, and what SmartVerify recorded but was not configured to refuse. It also states the share of traffic that could not be classified.

See SmartVerify on your own data path.

Review the evidence record, deployment model, and coverage for your environment.