Aleytheya
ἀλήθεια
The Infrastructure for Insurable AI
Aleytheya · Confidential
01 / 24
The Blackbox Problem.
“I do not fully understand how neural networks work internally, despite having spent my career building them.”123
GH
GEOFFREY HINTON
2024 Nobel Laureate, Physics · Nobel address, October 2024
Enterprise AI adoption5
Adoption has been rampant, hand-in-hand with technological advancement — adaptation has not kept up.
- AI agents are probabilistic models forced to make deterministic decisions.4
- A statistically significant non-zero failure rate is architecturally inevitable — not a bug to be patched, a property of the system class.
- Governance, insurance, and containment frameworks for AI failures remain in their infancy.
- Enterprises deploy these systems with the diligence of deterministic software — the mismatch creates institutional blind spots.
Three forces compound
- Capability racing ahead — the architects do not fully understand the architectures they ship.
- Adoption matches pace — nearly three-quarters of enterprises now run AI in production.
- Institutional response lags — no insurer underwrites this risk; no operational framework contains it.
This gap has produced a string of high-profile incidents. Four prominent examples shown in the next two slides.
Aleytheya · Confidential
02 / 24
When AI Fails.
Case I · Commitment
Air Canada Chatbot
- Chatbot fabricated a bereavement-fare refund policy for passenger Jake Moffatt — policy did not exist.
- The chatbot was Air Canada’s official customer-service interface, not a third-party tool.
- Airline argued the chatbot was “a separate legal entity”.
- First major airline forced to honour AI-fabricated terms in court.
Damages
C$ 812
Takeaway
AI agent made a binding commitment the company never authorised. The legal system held the company liable regardless — setting a precedent for every consumer-facing AI in production.
Case II · Confidentiality
ChatGPT Training-Data Extraction
- Researchers at Google DeepMind, ETH Zurich and UC Berkeley demonstrated: repeating a single word thousands of times caused ChatGPT to regurgitate verbatim training data.
- Leaked: real names, email addresses, phone numbers, private content scraped from the internet.
- The attack exploited a fundamental memorisation behaviour, not an implementation bug.
- OpenAI patched the specific vector — the underlying training-data leakage remains a class of risk.
Attack cost
< $ 200
Takeaway
Privacy is an architectural vulnerability, not a policy gap. No volume of contractual or audit controls eliminates the leakage — it must be addressed at the runtime layer, before training data can be reconstructed from model behaviour.
Aleytheya · Confidential
03 / 24
When AI Fails.
Case III · Containment
OpenClaw Deletes Meta Director’s Inbox
- Summer Yue (Meta’s AI alignment director) connected the OpenClaw AI agent to her primary inbox for email sorting.
- Agent operated under explicit user authorisation — the failure was scope creep, not intrusion.
- Root cause: context-window compaction caused the agent to forget its own safety instructions.1
- No technical safeguard prevented the loss — recovery depended on the email server’s own backups.
Emails deleted
200 +
Takeaway
Autonomous agents may lose safety constraints mid-operation. Once acting, no guaranteed kill switch — the agent that ignored ‘STOP’ is the same agent enterprises are wiring into production systems today.
Case IV · Discrimination
iTutorGroup Age Discrimination
- EEOC suit: AI hiring tool automatically rejected applicants above a certain age — women over 55, men over 60.
- First EEOC enforcement action against an AI hiring tool.
- Algorithm procured from a vendor without bias audit — liability still landed on the employer.
- Violated the Age Discrimination in Employment Act.
Settlement
$ 365,000
Takeaway
Competent professionals lost employment through no fault of their own. AI failure translated into tangible harm — and the employer bore the financial and regulatory consequences, not the vendor.
Aleytheya · Confidential
04 / 24
The Heavy-Tail Distribution.
For 99% of the time, AI works as designed. For the 1% it fails, it fails catastrophically — owing to issues including, but not limited to: organisations using previous versions of GPT to train the next1, training data not appropriately diversified or vetted2, safety instructions lost mid-operation due to context-window compaction3, and the OWASP Top 10 catalogue of LLM-specific risks — prompt injection, insecure output handling, training-data poisoning, model DoS, supply-chain vulnerabilities, and more.4
Impacts are tangible
Teachers denied employment, executives who lost data, consumers who relied on fabricated policies.
This is the tail end of the distribution — expected outcomes are benign, but extreme outcomes carry outsized financial and legal consequences.
90%+ of businesses report wanting coverage for AI risks, yet no insurer is purpose-built to price this category.5
The litigation accelerant
Every catastrophic outcome creates documentary evidence — log files, customer complaints, regulatory filings. Each becomes admissible against the deploying enterprise. The asymmetry is structural: damages compound while accountability frameworks lag.
Aleytheya · Confidential
05 / 24
YC Spring 2026 Companies.
We believe it was failures like these that led to the founding of two YC-backed companies (Spring 2026 cohort).
Klaimee
YC · P26
Founders
Ines Boutemadja CEO · ex-SafetyWing GM · HEC Paris / NUS
Julien Catonnet CTO
Julien Catonnet CTO
Mechanism
- Certifies and insures AI agents at the point of deployment.
- Runs structured evaluations in a sandbox against a generalised threat list.
- Sandbox findings produce a risk score, which determines coverage tier and ceiling.
- Risk score is rendered as a certification badge (a renewable artefact, not a runtime signal).
- Financial guarantee & liability coverage bound at the assessed risk level.
- Covers third-party damage and own damage (first-party).12
Team size
2
Mount Insurance
YC · P26
Founders
Fabian Amherd ETH Zürich · CS / ML
John Bachmann Co-founder
John Bachmann Co-founder
Mechanism
- Red-teams AI agents against simulated adversarial scenarios.
- Adversarial scenarios drawn from a fixed catalogue of known attack patterns.
- Findings produce an ADR certification — positioned as “SOC 2 for AI agents”.
- ADR is a periodic artefact: refreshed on evaluation cycles, not in real time.
- Underwrites coverage for direct financial loss based on red-team findings.
- Coverage: unauthorised actions, erroneous actions, data & tool misuse, manipulation events.34
Team size
2
Aleytheya · Confidential
06 / 24
An Antiquated Approach.
AIG · first cyber-insurance policy
1997
Cybersecurity expert visits company in person → manually inspects infrastructure & vulnerabilities → subjectively assesses the internal team → presents findings to underwriter → premium generated by hand.1
Manual. Arbitrary. Snapshot in time.
Klaimee and Mount use separate, equally outmoded mechanisms.
Klaimee
Runs structured evaluations in a sandboxed environment against a generalised threat list; generates a risk score and certification badge; binds a financial guarantee at the assessed risk level.2
Mount
Red-teams the deployed agent against simulated adversarial scenarios; issues an ADR certification (“SOC 2 for AI agents”); underwrites coverage based on red-team findings.3
The shared fundamental flaw
Testing is one-time. Coverage persists.
Zero accounting for model drift, model repurposing, evolving threats, or runtime behavioural changes. AI-agent risk profiles may change significantly within days of the simulation. There is a reason these passively calculated, one-time-assessed, snapshot-dependent, decay-prone, non-adaptive methodologies did not work in the cybersecurity space.
Aleytheya · Confidential
07 / 24
Industry Abhors a Vacuum.
It is one of the redeeming features of capitalism that if there is a gap in the market, players will emerge to capitalise on it. Status quo is not a moat one can rely on indefinitely. In late 2016 / early 2017, the concept of “active insurance” emerged in cybersecurity.
Coalition
Position
Pioneer of the active-insurance model. Built a security operations centre into the underwriting stack — every customer monitored continuously, claims data flowing back into risk models in real time. Largest cyber-insurance startup by GWP.
At-Bay
Approach
Continuous monitoring on client infrastructure
Position
Specialised early in ransomware — the heaviest claims category in cyber. Same active-monitoring principle: telemetry from client infrastructure flows back into pricing. The 7× edge is direct flywheel evidence.
Aleytheya · Confidential
08 / 24
Active Insurance.
How active insurance works
- Platform / operations centre deployed on client infrastructure.
- Proactively scans and eliminates threats in real time.
- Underlying loss distribution for underwriting derived from operations conducted in real life, in real time.
- Confidence in own scanning → ability to insure with a true premium.
- Premium tracks behaviour: when the system changes, the policy responds — not just the price.
Data flywheel effect
- More clients → more threat data → better detection → fewer claims → lower loss ratios → competitive premiums → more clients.
- Each rotation widens the gap between active insurer and passive incumbent.
- Flywheel compounds over time into an increasingly defensible moat.
Aleytheya · Confidential
09 / 24
Operational Convergence.
- Reverse-proxy layer through which all AI-agent requests and responses are routed.
- Deployable in minutes — no SDK changes, no agent rewrite, works with every major LLM provider.
- Scans actively for: prompt injection, PII leakage, destructive tool calls, hallucinated outputs, runaway loops, policy violations.
- Based on severity: flags / masks / blocks AI misoperations in real time.
- Telemetry is the underwriting input — every request, every response, every tool call feeds the model.
- Because we see every request and response in production → map AI misoperations to real financial data.
- Generates an expected-loss distribution as precise as actuarial models underpinning mature insurance lines.
- Pure premium computed directly from this distribution — expected loss + transparent risk load — not from a one-time sandbox simulation.
- Risk transfer is computed, not negotiated — the premium is a derivative of the live system.
- Cerberus produces the data. Icarus produces the price. Parachute issues the policy. One telemetry stream, three outputs.
This functioning has been primarily demarcated into three products:
Cerberus
+
Icarus Threshold
+
Icarus’ Parachute
Aleytheya · Confidential
10 / 24
The Cerberus Protocol.
- Sits between AI agents, tools, and LLM providers.
- Checks every request and response.
- Blocks unsafe actions when policy requires.
- Builds an evidence trail for compliance.
- No SDK changes, no agent rewrite, works with every major LLM provider.1
Core principle
Passive until risk appears. Active when it matters.
The Cerberus metaphor
Greek three-headed guardian of the underworld — a domain everyone imagines but none can truly see, like a neural network. Three heads → Secure · Contain · Protect.
Cerberus
GUARDIAN OF THE UNDERWORLD
SCP — reference I
Named after the SCP Foundation, a wiki cataloguing anomalies too dangerous for public exposure. Each anomaly must be Secured (prevent triggering), Contained (limit blast radius), and Protected against (shield the wider world).
SCP — reference II
Containment procedures transfer between structurally related anomalies — institutional learning compounds. Same principle at Aleytheya: every AI agent is an anomaly. The system learns from every contained incident.
SCP Foundation
Aleytheya · Confidential
11 / 24
Secure. Contain. Protect.1
Secure
I · INBOUND
Block adversarial inputs before they reach the model.
- Prompt-injection defence. Catches jailbreaks, role overrides, system-prompt extraction, and indirect injection delivered via uploaded documents, RAG sources, or pasted email content — the dominant attack surface for production agents.
- Sensitive-data scanning. PII detection across US, EU, India, UK, Canada and Australia formats — SSN, Aadhaar, NINO, SIN, TFN, credit cards, bank IDs, passports, medical IDs — with locale-aware regex and ML-backed classifiers.
- Spend caps + runaway-loop detection. Configurable budgets per-call, per-day and per-month, with a hard ceiling of 60 calls per minute that triggers automatic throttle before runaway loops compound into incidents.
- AZ-5 kill switch. One toggle halts every agent under the proxy immediately — reversible, audited, designed for the moment when human override is the only option that scales.
Contain
II · OUTBOUND
Keep model actions inside approved boundaries.
- Destructive-action blocking. DROP / TRUNCATE on databases,
rm -rf, unauthorised payments, mass communications, IAM escalation and bulk data exfiltration — intercepted before the tool call reaches the executor. - Tool allow-listing. Per-agent allow-lists with parameter validation and anomalous-tool detection — an agent calling a tool it has never used before is flagged before the call executes.
- Regulated-content flagging. Medical, legal, financial and discriminatory outputs intercepted and escalated according to jurisdiction-specific rules — configurable per business unit and per geography.
- Drift detection. Per-agent baselines for tool selection, topic distribution and language — behavioural deviations from baseline trigger re-evaluation of risk class before they become claims.
Protect
III · COMPLIANCE
Build the tamper-proof record regulators require.
- Merkle-chained audit trail. Append-only, 5-year retention, no deletion API — every entry cryptographically linked to the next, so any post-hoc tampering is provably detectable.
- 11 compliance frameworks. EU AI Act, GDPR, SOC 2 Type II, HIPAA, ISO 27001, ISO 42001, PCI-DSS, NIST AI RMF, DPDPA (India), Colorado AI Act and SOX — one runtime layer satisfies the evidence requirements of all eleven.
- Auto-disclaimers + trace stripping + language enforcement. Outputs are sanitised before they leave the proxy — PII redacted, internal trace IDs removed, mandated disclaimers attached in the regulated language of the deployment jurisdiction.
- Data residency by architecture. AES-256-GCM at rest, metadata-only by default — Aleytheya never sees prompts or responses unless the customer explicitly opts in, region by region.
Aleytheya · Confidential
12 / 24
The Icarus Threshold.
The Icarus Threshold measures the distance between ambition and catastrophe.
- Continuously transforms runtime events, agent behaviour and operational signals → financial exposure ranges.
- Beyond compliance theatre and meaningless risk scores: quantified operational exposure in dollars.
- Risk expressed in monetary terms is interrogable by boards, brokers and regulators alike — the universal language of consequence.
- Eliminates the certification-coverage gap: exposure tracks the actual deployment, not its description at sign-off.
- Defensible inputs — industry priors, customer telemetry, transparent rationale — produce a continuous, policy-ready signal, not a one-off report.
How it works · Stages of the Icarus Threshold
01
OBSERVE
Telemetry
Every agent event → structured signals: model, tool, latency, cost, policy result, risk flag.
▼
02
PROFILE
Agent profile
Auto-profiles each agent from observed traffic — domain, autonomy, tools, data sensitivity. Draft → confirm in 30 s. Drift detection. Quarterly reconfirmation.
▼
03
MODEL
Risk intelligence
Continuously transforms runtime telemetry → operational and financial risk intelligence. Industry priors → customer-calibrated → transparent inputs.
▼
04
REPORT
Dollar ranges
Dollar range, not coloured flags. Quarterly exposure ranges — high / low band, 90-day rolling, per business unit. Board-ready: Policy → Rule → Evidence chain.
Aleytheya · Confidential1
13 / 24
A Parachute for Icarus.
The myth of Icarus was not a warning against flight. It was a warning against believing altitude removes consequence.
Not a quote pulled from a vendor’s gut. A premium derived from your loss distribution — event by event.
How the premium is built
01
Inherit
Start from the Icarus Threshold loss distribution. The same telemetry that produces dollar exposure ranges becomes the basis for the premium — no separate sandbox, no separate dataset.
→
02
Integrate
Actuarially fair base premium derived directly from the distribution — pure mathematics, no opinions. The expectation of aggregate loss, computed continuously from production data.
→
03
Profile
Layer in AI-specific characteristics: autonomy level, data class, tool surface, operator maturity. Each modifier is auditable and traceable to a documented profile dimension.
04
Adjust
Transparent debits and credits applied for behaviour, control quality and operational maturity — visible, defensible rationale on every premium adjustment, line item by line item.
→
05
Quote
Final premium an underwriter, broker or captive can stand behind line by line — defensible against regulator scrutiny, broker challenge and reinsurer review. Produced as an auditable document, not a marketing number.
→
06
Recalibrate
Premium re-prices at a frequency of your choosing as behaviour shifts — never stale. Daily, weekly, quarterly — the cadence is yours; the model runs continuously underneath.
Aleytheya · Confidential1
14 / 24
Aleytheya Chat.
In 2026, we have reached the agentic era. People no longer want dashboards. All they care about is answers.
- Sits on top of every event, policy and risk signal in the platform.
- Ask in plain English → get answers with events, dollar amounts, policy references.
- Risk in plain English — dollar numbers, events, next actions.
- Evidence on demand — every answer cites Merkle-chained audit records.
- Question to action — draft the policy, propose the rule, route for approval. One confirm. No YAML, no tickets.
- Connect any LLM without restriction.
- MCP server support planned (any environment, anywhere); currently on website.1
Tagline
No technical expertise needed.
Talk with your agents’ data.
Talk with your agents’ data.
Aleytheya · Confidential
15 / 24
The Guardrails Gold Rush.
Aleytheya is not just an insurance company. Cerberus alone positions us in the AI guardrails & governance sector — a market that is lucrative, evolving, and in extraordinary demand.
Lakera
Acquisition
~$300M
Revenue
~$5.7M
Multiple
~52×
The pattern
Security incumbents — Check Point ($20B+ market cap) and Palo Alto Networks ($130B+ market cap) — paying 44–52× revenue for AI-guardrails companies with sub-$20M ARR. Cerberus competes directly in this space.
Aleytheya advantage
We provide a calibrated dollar range and individual risk metrics — not a custom risk score that isn’t interpretable.
Aleytheya · Confidential
17 / 24
Cerberus vs the Competition.
| Capability | Fiddler.AI | Arize AI | Aleytheya · Cerberus |
|---|---|---|---|
| Architecture | Trust Service — per-call dependency on hot path (5M req/day vendor-lock concern)1 | Reactive post-call (reask / fix / filter)2 | Inline reverse proxy on the critical path. |
| Hard inline block | Limited — policy via Trust Service hooks1 | ✗ no on_fail="block" option exists2 |
✓ blocks on critical path, every time. |
| Hot-path overhead | Per-call dependency on external Trust Service1 | Out-of-band post-call — no live enforcement | Single proxy hop, asynchronously offloaded heavy checks. |
| Coverage breadth | Input metrics + post-call evals8 | Traces, evals, drift — no tool-call interception2 | Input + output + tool calls + telemetry, every request. |
| Deployment friction | pip install + Python init in Jupyter; 3–5 lines of HTTP per guardrail3 | ML-engineer focus; G2 reviews flag steep learning curve4 | GUI + YAML — no code required, no agent rewrite. |
| Vendor independence | Trust Service hot-path lock-in1 | Phoenix OSS available; AX features (Copilot, dashboards, HIPAA) gated7 | ✓ fully vendor-neutral, no lock-in. |
| Pricing accessibility | Annual commitment, no permanent free tier5 | Free tier restricted — spans cut 100K → 50K, overage pricing removed6 | Planned self-serve + SMB tier. |
| Compliance frameworks | Metrics: drift, toxicity, jailbreak, faithfulness8 | Traces, evals, drift2 | ✓ 11 frameworks + Merkle-chained audit. |
| $-quantified exposure | ✗ never produces $ output8 | ✗ surfaces traces/evals/drift, never $2 | ✓ Icarus Threshold — live dollar bands. |
| Data residency | Not specified | Not specified | ✓ metadata-only, customer-controlled storage, AES-256-GCM. |
| Capital position | ~$100M total — modest against $20B+/$130B+ incumbents11 | ~$131M total | Pre-seed, lean. |
Cerberus is the only entry in this table that combines inline runtime enforcement, full-surface coverage (input + output + tool calls), vendor neutrality and a dollar-denominated exposure layer on top.
Aleytheya · Confidential
18 / 24
Icarus vs the Competition.
| Capability | Klaimee | Mount | Aleytheya · Icarus |
|---|---|---|---|
| Risk-assessment method | One-time sandboxed simulation1 | One-time red-team evaluation2 | ✓ Continuous production monitoring. |
| Loss-distribution estimation | ✗ categorical risk score, no distribution1 | ✗ categorical risk score, no distribution2 | ✓ Compound-Poisson aggregate-loss distribution, recalibrated event-by-event. |
| Premium calibration | Static — set at certification time1 | Static — set at evaluation time2 | ✓ Continuous, behaviour-linked, re-priced at a frequency of your choosing. |
| Actuarial framework | None — certification tier as proxy1 | None — ADR certificate as proxy2 | ✓ Pure premium = E[L] + risk load + agent-specific debits / credits. |
| Continuous Bayesian updating | ✗ | ✗ | ✓ Bühlmann–Straub credibility model. |
| Model-drift accounting | ✗ risk profile frozen at certification | ✗ risk profile frozen at red-team date | ✓ Per-agent baseline tracking + drift triggers re-evaluation. |
| Data flywheel | ✗ sandbox-only data, no production loop | ✗ simulation-only data, no production loop | ✓ Every production request / response feeds the next premium. |
| Reserve calibration | Held against certification ceilings | Held against ADR-defined event categories | ✓ Held against live exposure — capital efficient at portfolio scale. |
| Coverage scope | 1st party + 3rd party at the certified tier1 | Direct financial loss within event catalogue2 | ✓ Configurable per agent class, per business unit, per event family. |
| Compliance / audit | Certification badge | ADR certification | ✓ Merkle-chained audit trail, 11 frameworks. |
| Methodology era | 1997 — passive, one-time assessment | 1997 — passive, one-time assessment | 2017+ — active, continuous, actuarially grounded. |
Aleytheya is the only entry with a continuous actuarial framework, a production data flywheel, and a runtime enforcement layer feeding the underwriting model. Klaimee and Mount are categorical risk-rating engines with a guarantee bolted on. Icarus is an insurance company.
Aleytheya · Confidential
19 / 24
The Team.
Co-founder · CTO
Vaibhav Chopra
SDE 1 · Atlan
Now · Atlan
—Defining the agentic infrastructure behind app & connector development; building a self-evolving SDK.
—Optimized real-time data pipelines for JPMC, Nasdaq & Autodesk — +80% performance.
—Maintains Atlan’s Python SDK (250K+ downloads/mo); built data-lineage & governance packages.
—Defined an AI-driven SDLC — connector builds from weeks to days via reusable scaffolding.
Earlier · Agmitra Technologies — SWE Intern
—Cut polarimeter data-collection 71% (2 min → 35 s); REST APIs <2 s — −65% server load, +30% load speed.
Recognition
Finalist, NFTE World Series of Innovation · Gold Medal, International Information Olympiad · UN MCN Emerging Technologist Fellow.
Co-founder · CEO
Chanakya Rao
ML Engineer · Policybazaar.com
Now · Policybazaar
—Re-architected the bank-statement OCR pipeline with multiprocessing & batch-level parallelism.
—Refactored a legacy policy-document pipeline — sharper table detection & layout-aware extraction from semi-structured PDFs — after a customer escalated an AI-rejected claim to IRDAI, India’s insurance regulator.
Earlier · Loadshare Networks — Data Analyst Intern
—Reverse-engineered route polylines; via survival analysis surfaced latent rest-break distributions as the dominant source of delivery-time dispersion — launching rider rest-signaling with real-time customer push notifications.
Recognition
—IEEE paper — ML models deployable on resource-constrained drones to detect crowd density & mob violence.
—IEEE paper — an ultra-low-cost IoT system that drops into any switchboard, using a sensor network to drive lighting & HVAC.
Both UN Millennium Campus Network Fellows — selected from 44,000+ applicants — and each chosen as Campus Director (from 4,000+ entries) to lead the fellowship’s leadership sessions.
Aleytheya · Confidential
20 / 24
Pricing Model.
SaaS metered per-event by tier — Parachute premiums priced actuarially per agent from live telemetry.
10 frameworks
SOC 2 · ISO 27001 · ISO 27701 · NIST AI RMF · EU AI Act · HIPAA · PCI DSS · GDPR · GLBA · DPDPA
Starter
$499
/ month
Up to 250k events / mo
$2.20 / 1k overage
$2.20 / 1k overage
Cerberus full · Chat · proxy
Most common
Govern
$2,500
/ month
Up to 2M events / mo
$1.25 / 1k overage
$1.25 / 1k overage
+ Icarus Threshold · audit trail · SSO-ready
Insured ↓
Risk
$8,000
/ month
Up to 10M events / mo
$1.00 / 1k overage
$1.00 / 1k overage
+ Parachute eligible · production SLAs
Enterprise
$12,000
/ month
Up to 20M events / mo
$0.90 / 1k overage
$0.90 / 1k overage
+ Executive Briefing · custom policies
Free · $0
0–10k events / mo · hard cap · Cerberus basic + Chat
Strategic · contact
20M+ events · private deployment · carrier workflows
Parachute · how the premium is priced (per agent)
premium = expected loss × carrier load × event volume
Expected loss
Incident frequency × severity, drawn live from Cerberus telemetry and the Icarus loss distribution. Autonomy, PII exposure, and tool blast-radius drive the curve.
Carrier load
Capital reserves and reinsurance against tail events. Set by Aleytheya’s carrier partners; reflects the cost of warehousing the risk.
Event volume
Billable agent activity per month. Higher volume → more exposure → higher premium — same axis the SaaS tiers are metered on.
Live range
$40 – $850
per agent / mo · median $180
Aleytheya · Confidential
21 / 24
The Path to $100M ARR.
$89.4M SaaS + $10.3M insurance-linked by Month 24 — ~33% month-over-month compounding off a Month-12 base of $5.3M.
ARR by month · M2 → M24
ARR added per 6 months · compounding
M2–M7
+$0.9M
M8–M13
+$6.3M
M14–M19
+$31.2M
M20–M24
+$61.3M
Total ARR · M24
$99.7M
$89.4M SAAS$10.3M INS
Customers · M24
2,049
paid SaaS accounts · + 187 insured
Govern $35.4M (1,180) · Risk $34.4M (358)
Enterprise $12.7M (88) · Strategic $4.4M (11) · Starter $2.5M (412)
Enterprise $12.7M (88) · Strategic $4.4M (11) · Starter $2.5M (412)
Op profit run-rate · M24
$2.9M / mo
~35% op margin · 5.4B events / mo
Aleytheya · Confidential
22 / 24
The Ask.
$1.3M pre-seed buys 18 months of runway, SOC 2 readiness, and a $1M ARR proof-point before Series A.
Use of funds · $1.3M pre-seed
PROD+ENG · 35%
GTM · 20%
COMPLIANCE · 18%
CLOUD · 15%
TOOLS · 8%
4%
$450k
$260k
$235k
$195k
$105k
$55k
Product + engineering
GTM + outreach
Compliance + legal
Cloud + AI + data
Tools + admin
Contingency
$1.3M
Pre-seed · 18 months runway
Bridges to Series A at $1M ARR · M18.
Milestones · pre-seed to Series A
M3
Design partners
Five anchor customers under LOI — two regulated, three high-autonomy AI.
→
M6
SOC 2 ready
Type I attestation in flight; ISO 27001 readiness assessment complete; CISO retained.
→
M12
$1M ARR
50 paid customers across Govern + Risk tiers; Cerberus + Icarus in production.
→
M18
Series A ready
Parachute pilot live with one carrier; unit economics proven on cohort data.
What the $235k compliance line buys
- Certifications: SOC 2 Type I/II, ISO/IEC 27001 (ISMS), ISO/IEC 27701 (privacy); GDPR + India DPDP alignment.
- Accredited CISO + external security auditor — validates the control architecture for enterprise procurement.
- CA / CFA review of loss-modelling — assumptions, confidence intervals, dollar-weighted adjustments. Financial defensibility for Parachute underwriting.
Two-year economics
Year 1 cumulative
$3.9M
$1.3M ASK + REVENUE + SERIES A
Year 2 cumulative
$47M
EXIT-MONTH RUN-RATE ~$5.4M/MO
Break-even
~$65M ARR · Month 21
Aleytheya · Confidential
23 / 24
Aleytheya
Thank you for your time
and consideration.
We look forward to building this with you.
Aleytheya · Confidential
24 / 24