Silmaril

Prompt injection defense that is self-healing

Spring 2026ActiveB2BSecuritySan Francisco, CA, USA

Company

Silmaril is self-healing prompt injection defense for AI native applications and agents. It understands application context to block 2x as many threats as current SOTA defenses with 10x lower latency. Customers plug us into agentic frameworks like LangGraph with 5 lines of code. Silmaril has stopped $28M of damages for customers.

Note: This is a preliminary assessment based on limited publicly available information. We did not have access to LinkedIn profiles or live product screenshots for this analysis. We will update this entry with a more thorough review soon.

Verdict

High Signal

Market Opportunity

AI-native applications and agentic frameworks are exploding in enterprise adoption, making prompt injection a tier-1 security threat. The ICP is clear — companies building on LangGraph, AutoGPT-style frameworks, and enterprise LLM pipelines. Security spend in this category will be multi-billion as AI agents handle sensitive operations; this is B2B enterprise with clear ROI-based selling.

High Signal

Founder Signal

Aum built the security and privacy framework at AWS preventing $1.8B in damages — directly relevant pedigree. Eduardo is an ex-Amazon tech lead who built low-latency ML models generating $400M annual revenue, and claims to have chained a prompt injection into root access inside ChatGPT, demonstrating hands-on offensive security credibility. Both founders have deep, relevant AWS/Amazon backgrounds with quantified impact. LinkedIn not available, so claims can't be fully verified, but the specificity is a positive signal.

Medium Signal

Competition

No competitor data returned, but this space includes Rebuff, LLM Guard, Lakera Guard, and broader AI security platforms like Protect AI and HiddenLayer. Silmaril's differentiation claim — self-healing, context-aware defense with lower latency — is meaningful if real. The risk is that major cloud providers (AWS, Azure, Google) could bundle similar protections natively, and the category is attracting VC-backed players rapidly.

Medium Signal

Product

Claims are specific and compelling — $28M in damages stopped, 2x better than SOTA, 10x lower latency, 5-line integration with LangGraph. However, no named customer logos, no pricing page, no public demo or API docs visible. The metrics are precise enough to suggest real customers but unverifiable from available data.

OverallA Tier

Two ex-Amazon founders with deep, quantified security and ML backgrounds attacking a real and urgent enterprise problem — prompt injection is not a niche concern, it's the primary attack surface for agentic AI systems. The $28M damages-stopped claim and 2x/10x performance metrics suggest real traction, not vaporware. The main weaknesses are unverifiable founder LinkedIn data and no visible named customers or public product artifacts. Competition from cloud providers bundling security is a real long-term risk. If the technical claims hold up under scrutiny, this is a strong team in a high-urgency market with a defensible technical approach.

Active Founders

Aum Upadhyay

Founder

Co-Founder & CEO at Silmaril. I built the security and privacy framework at AWS that prevented over $1.8B in damages. Now I bring that same defensive mindset to Silmaril, working with customers to ship prompt injection defense.

Eduardo Velasco

Founder

Co-Founder & CTO at Silmaril, the world's first self-healing prompt injection defense. I prove the risks are real by finding exploits myself. I chained a prompt injection into root access inside ChatGPT. Ex-Amazon tech lead specializing in low latency ML models that generated $400M in annual revenue.

Silmaril

TierA Tier

BatchSpring 2026

Team Size2

StatusActive

LocationSan Francisco, CA, USA

Disagree with this rating? Let us know

$200 GPU creditsClaim yours →