Product Scope After the Acquisitions
Lakera should be evaluated as a Check Point product surface, not as a standalone AI firewall with an unchanged startup roadmap. Check Point announced the acquisition agreement on September 16, 2025 and reported closing it on October 22, 2025. The current Lakera documentation is titled the Check Point AI Defense Plane and divides coverage into Workforce AI Security, AI Agent Security, AI Guardrails, and AI Red Teaming. The products share agent inventory across workforce and agent-security views, giving the buyer a path from discovery and governance to runtime blocking and adversarial assessment.
Prompt Security is also acquisition-owned, but its current public surface remains more focused. SentinelOne announced the deal on August 5, 2025, and its SEC filing states the transaction completed on September 5, 2025. Prompt's official site still organizes the product around employees, homegrown GenAI applications, AI code assistants, agentic AI/MCP, and red teaming. That continuity matters: buyers can evaluate concrete Prompt workflows today, but should confirm how licensing, telemetry and administration connect to the wider SentinelOne Singularity platform rather than assuming the announcement created a single bundled SKU.
Runtime Protection for Applications and Agents
Check Point AI Guardrails, the current name for the Lakera Guard runtime layer, screens AI interactions for prompt attacks, data leakage, content violations and off-policy agent behavior through the Guard API. Current docs state that AI Agent Security discovers agents across platforms and cloud infrastructure, inventories tools and MCP servers, assigns per-agent risk ratings, and maps risk types to OWASP and MITRE ATLAS. AI Guardrails can also be purchased as a standalone tier, so a team can start at the application boundary without adopting every Check Point AI security component on day one.
Prompt Security protects both sides of an AI interaction: the official homegrown-app material describes inspection of inputs and outputs for prompt injection, jailbreaks, denial-of-wallet patterns, data leaks and toxic content, while its workforce material focuses on monitoring and enforcing employee use. This breadth is meaningful, but its center of gravity is policy enforcement wherever people, applications and agents touch AI. Lakera wins the default runtime decision because its current architecture combines interaction-level protection with a documented agent inventory and posture layer rather than treating every surface as an isolated policy channel.
Workforce, Code Assistant, and MCP Governance
Lakera's Check Point documentation separates Workforce AI Security into Enterprise and Essentials options. Enterprise covers discovery, governance and protection across web, desktop and developer tools; Essentials is web-only. Administrators deploy endpoint components through existing device-management tools, and the platform evaluates application context, content sensitivity and user behavior to block risky activity and preserve an audit trail. That is a full workforce program, not only an API filter, and it corrects older descriptions that framed Lakera exclusively as middleware in front of an LLM.
Prompt Security remains especially credible when workforce AI and MCP are the first control points. Its official pages cover shadow-AI discovery, automatic anonymization or blocking of sensitive data, department/user policies, code-assistant governance, and an MCP Gateway that inspects requests and responses between agents and servers. The agentic page also describes MCP discovery and risk scoring across more than 13,000 known servers; that number is a vendor-published inventory claim, not an aicoolies benchmark. Prompt is the better specialist for a narrowly scoped MCP or employee-use rollout, while Lakera remains the broader program winner.
Deployment, Data Handling, and Operations
Lakera's current docs state that AI Agent Security is enterprise SaaS and that AI Guardrails can be self-hosted/on-prem for tighter compliance control. The official pricing application adds concrete operational distinctions: Community is $0 for 10,000 requests per month on SaaS, while Enterprise is quote-based with a flexible request allowance and SaaS or self-hosted deployment. The docs also list model compatibility across hosted, open-source, custom and fine-tuned models. These details make deployment and request capacity explicit procurement axes rather than generic security promises.
Prompt's homegrown-app material references API and on-premise patterns, while the employee product depends on coverage across user interaction surfaces. Its acquisition announcement describes browser, desktop and API integration plus model-agnostic coverage, but announcement language must be checked against the current quote and technical design before purchase. A buyer should document where prompts and outputs are inspected, whether traffic is proxied or endpoint-observed, which MCP calls are intercepted, where logs are retained, and how policies fail when a component is unavailable. No deployment should be described as frictionless without a real pilot.
Pricing and Procurement Boundaries
Lakera's current official pricing app lists Community at $0 for 10,000 requests per month on SaaS. Enterprise is quote-based, uses a flexible request allowance, and supports SaaS or self-hosted deployment. Those values replace the stale tier structure previously stored in the aicoolies base record. Request volume alone is not the full comparison: support, governance scope, hosting requirements and the agent, workforce or red-team modules included in a quote can outweigh the headline allowance.
Prompt Security's current public product pages route buyers to a demo rather than publishing a numerical self-serve pricing grid. The honest comparison is therefore quote-to-quote: count protected employees, applications, code-assistant users, agent/MCP interactions, deployment components, retention, support and any SentinelOne platform prerequisites. Prompt may produce a cleaner focused quote for one governance surface, but that is a procurement hypothesis to test. Lakera wins because its public entry tier, runtime allowance and deployment choices are clearer while its broader architecture reduces the number of disconnected AI security evaluations.
Final Verdict: Lakera for the Broader Program
Choose Lakera when the security roadmap includes AI applications, autonomous agents, workforce AI use and adversarial assessment under one control model. Its current Check Point architecture connects agent discovery, risk ratings, runtime Guard API controls and workforce inventory; the standalone AI Guardrails tier still supports a narrower first deployment. This does not prove higher detection accuracy or lower latency, and no such benchmark was performed. It makes Lakera the stronger default because the documented product boundary matches the full lifecycle that enterprise AI teams now need to govern.