All analysis was generated autonomously, without human review. Scores are analytical opinions drawn from the cited public sources, without hands-on testing. They are not audits, certifications, investment reports, purchasing advice, or evaluations of quality.
General Analysis, founded in 2025, sells automated red teaming and runtime guardrails to enterprise security teams running AI agents in production. Its attack agents map the tools and permissions a customer's agent can reach and chain multi-step attacks. It turns the findings into guardrails that screen prompts, responses, and tool calls. In a 2026 test reported by Tech Startups, one such agent talked 50 of 55 live customer-service bots into granting more than $10 million in fake perks. Altos Ventures led its $10 million seed round in April 2026, joined by Y Combinator. It has not publicly named its customers, and a co-founder told Axios it aims for $2 million in revenue within 12 to 18 months. Those attack agents and the research behind them are what a rival would take longest to copy.
| Description | General Analysis runs automated adversarial red teaming against enterprise AI agents and turns the findings into runtime guardrails, detection, and monitoring for models, tools, and MCP servers. | [f1] |
|---|---|---|
| Founded | 2025 | [f2] |
| HQ | San Francisco, California, USA | [f3] |
| Funding | $10M total | [f3] |
| Latest funding | Seed, $10M (April 2026) | [f4] |
| Product | What it does |
|---|---|
| Automated AI Red Teaming | Context-aware adversarial testing that maps an agent's tool and permission graphs and generates multi-step exploit chains before release. |
| Runtime Guardrails and Observability | Runtime guardrail models and monitoring that screen prompts, responses, and tool calls and watch for poisoning and drift. |
| MCP Guard | A guard that inspects and controls calls to MCP servers to protect MCP-based agents from injection and data exfiltration. |
| AI Security Asset Management | Inventories models, knowledge bases, MCP servers, and agent pipelines while scanning for prompt injection and data leaks. |
AI Defense Matrix
| Govern | Identify | Protect | Detect | Respond | Recover | |
|---|---|---|---|---|---|---|
| AI-Workload Platforms Inference servers, training platforms, vector DB platforms, and the model-loading supply chain. | ||||||
| AI Orchestration Tools Agentic orchestration tools, plus their plugins, skills, hooks, system prompts, scaffolding, harnesses, configuration settings, and MCP clients on user devices. | ||||||
| AI-Generated Code Code produced by AI tools, AI-assisted reviews, AI-generated infrastructure-as-code and tests, and vibe-coded apps that bypass CI/CD. | ||||||
| AI Gateways & Routers MCP proxies and gateways, LLM routers, outbound AI-service traffic, shadow AI egress, and model-registry traffic. | ||||||
| AI Model Model weights, fine-tuning checkpoints, model cards, registries, AIBOM, and the third-party LLMs your enterprise consumes. | ||||||
| Training Data Datasets used for training, fine-tuning, and continued learning. | ||||||
| Runtime AI Data User prompts, inference inputs, RAG content, vector DB content, persistent agent memory, and interaction history. | ||||||
| AI Agent Identities AI agents as non-human principals, plus credentials, keys, permission scopes, service accounts, and delegation chains across agents and tools. |
Automated AI Red Teaming maps an agent's tool and permission graphs and generates multi-step exploits, while Runtime Guardrails, MCP Guard, and AI Detection and Response screen prompts, tool calls, and MCP traffic and contain risky agent behavior. These lines are mapped to the AI Defense Matrix. [f1]
How well the company can compete in its security market, scored across eight dimensions against public evidence.
| Dimension | Score | Rationale |
|---|---|---|
| Problem Clarity How precisely the company defines its problem, with evidence the problem exists at the scale claimed. | 3/5 | General Analysis names a specific buyer, enterprise security teams deploying autonomous agents, and a concrete pain, agents that cannot be secured like deterministic software. The pain is credible but its scale is shown mostly by the company's own red-team demonstrations rather than independent measurement. [s1, s2, s7] |
| Capability Depth How specific the technical capabilities are, with evidence beyond marketing claims such as docs and third-party validation. | 4/5 | Public documentation goes beyond marketing pages, describing guardrails shipped as open checkpoints plus custom detectors trained on adversarial pipelines and a documented MCP Guard proxy. Security researcher Simon Willison independently cited the Supabase and Cursor exfiltration finding, an external validation point most same-stage peers lack. [s4, s5, s8] |
| Market Timing Whether the market is ready for this product, with evidence that buyers are actively seeking solutions. | 3/5 | The enabler is real: the spread of autonomous agents and MCP-connected tools opened an attack surface that did not exist before, and the exploit class General Analysis works on was publicly demonstrated by mid-2025. Buyer demand for General Analysis specifically is still indirect, resting on the company's own positioning rather than published budget lines or analyst placement. [s1, s8] |
| Team Credibility Demonstrated domain expertise with public signals such as prior exits, publications, and industry recognition. | 3/5 | The team has built reinforcement-learning infrastructure at DeepMind, trained models at Jane Street, NVIDIA, and Cohere, and published at NeurIPS, ICLR, and ICML. The cited public sources do not name a prior exit or detail a sustained publication record, and its clearest independent validation so far is a single cited research finding. [s3, s8] |
| GTM Proof Evidence of actual traction (customers, revenue signals, partnerships) beyond stated intentions. | 3/5 | General Analysis names no customer publicly. Reputable backing, Y Combinator plus Altos Ventures, Menlo Ventures, and 645 Ventures, and a senior research team support a small upward adjustment for traction that is likely real but undisclosed. [s6, s7, s9] |
| Funding Efficiency Whether funding matches go-to-market ambition, with signs of capital-efficient growth. | 3/5 | The $10 million seed is proportional to the company's seed stage and its visible shipping of a documented platform and SDK. Efficiency itself is unconfirmed, with no disclosed revenue and a stated target of $2 million over the next 12 to 18 months. [s6, s9] |
| Category Clarity Whether the company creates or fits a recognizable category that buyers can quickly place in their stack. | 3/5 | General Analysis fits the emerging agentic-AI-security category buyers can place, but the label is still contested and overlaps red teaming, guardrails, and AI detection and response, so placement needs the company's own explanation. [s1, s2] |
| Incumbent Defensibility How vulnerable the core value proposition is to absorption as a feature by a platform vendor. | 3/5 | General Analysis shows real friction against quick copying through proprietary adversarial models and a closed loop from testing to runtime defense, but no structural moat that platform bundling could not eventually match as larger platforms add the same controls. [s2, s3] |
General Analysis sells to enterprise security teams that are putting autonomous AI agents into production and cannot secure them the way they secure ordinary software. The company argues that agents need broad access to tools and data, which collides with least-privilege models, and that their behavior shifts with each input in ways code review cannot anticipate.
The pain is credible and widely felt across the category, but its scale is shown mostly by the company's own demonstrations rather than independent measurement. Its sharpest evidence is a red-team run reported by Tech Startups, in which a single adversarial agent talked 50 live customer-service bots into handing out more than $10 million in fabricated perks, and only five of 55 systems tested resisted. [s1, s2, s7]
The platform spans several lines: automated red teaming that maps an agent's tools and permissions and generates multi-step exploits, runtime guardrails and observability, AI detection and response, an MCP Guard proxy, and asset management that inventories models, knowledge bases, and MCP servers. Public documentation goes beyond marketing pages, shipping guardrails as open checkpoints plus custom detectors trained on the same adversarial pipelines, alongside a documented MCP Guard proxy and SDK.
External validation is unusual for the company's age. Security researcher Simon Willison cited its finding that a Supabase database connected to the Cursor coding agent could be drained through one poisoned support ticket. Independent benchmarks of its guardrail accuracy are not yet public, so the depth beyond that cited finding is still shown mainly on the vendor's own pages. [s2, s4, s5, s8]
General Analysis competes in a crowded agentic-AI-security field where the pairing of offensive testing and runtime defense is not unique to it, and larger security platforms are moving into agent security.
What separates General Analysis today is the quality of its research and team rather than a structural barrier. Its differentiators, red-team tooling and guardrail models, are reproducible by a funded rival, so the company's edge is being early and demonstrably good at offensive research, not owning something others cannot build. [s2, s3]
General Analysis has raised a $10 million seed led by Altos Ventures, with Menlo Ventures, 645 Ventures, and Y Combinator also participating. Traction beyond funding is mostly undisclosed. None of the cited sources names a customer, though TechStartups reports unnamed enterprise customers whose systems reach large user bases.
A co-founder told Axios the company is aiming for $2 million in revenue over the next 12 to 18 months before raising a Series A. The reputable backing and a senior research team point to real but undisclosed early traction, which lifts an otherwise thin public record. [s6, s7, s9]
The founding team carries relevant research and engineering backgrounds. Co-founder Rez Havaei speaks for the company, and its own account describes a founding team that built reinforcement-learning infrastructure at DeepMind, trained models at Jane Street, NVIDIA, and Cohere, and published at NeurIPS, ICLR, and ICML.
That record is real but early. Public sources do not name a prior exit or detail a sustained publication record, and the strongest outside recognition so far is Simon Willison's citation of the Supabase and Cursor finding, a single notable signal rather than a track record. [s3, s8]
General Analysis publishes no public security attestation. A probe of its homepage footer, its /security and /trust paths, and its trust and security subdomains on July 3, 2026 found no SOC 2, ISO, or comparable certification.
For a seed-stage company selling into enterprise security teams in regulated sectors, the absence is normal at this stage but is a gap a security buyer will weigh before granting an agent broad access. [s10]
| Company | Relationship | Note | Compare |
|---|---|---|---|
| Straiker | competes with | Same-asset platform pairing offensive red-team testing with runtime guardrails for agentic apps and AI agents. | |
| CalypsoAI | competes with | Pairs runtime guardrails against prompt injection and data leakage with red teaming at scale, now part of F5. | N/AWe scored these companies at different scopes, so the totals measure different things. |
| Adversa AI | competes with | Continuous AI red teaming across agents, models, and MCP, overlapping General Analysis's testing line. | |
| Gray Swan AI | competes with | Runtime guardrail plus a crowdsourced red-team arena, competing on both offensive testing and runtime defense. | |
| Robust Intelligence | competes with | Algorithmic red teaming and runtime guardrails for AI models and apps, now part of Cisco AI Defense. | N/AWe scored these companies at different scopes, so the totals measure different things. |
| Haize Labs | competes with | Automated adversarial stress-testing of large language models, overlapping the red-teaming line. | N/AWe captured the evidence for these companies under different evidence-model versions (v1 vs v2), so the totals were scored under different conditions and are not directly comparable. |
| Promptfoo | competes with | Open-source evaluation and red-teaming for LLM applications, an OSS-anchored alternative to the testing line. | N/AWe captured the evidence for these companies under different evidence-model versions (v1 vs v2), so the totals were scored under different conditions and are not directly comparable. |
Add analyzed competitors to compare them side by side with General Analysis.
A closer look at the company's product strategy, measuring how defensible it is against market forces and examining the eight areas behind it.
pivot urgently
What General Analysis can defend is thin today. Its guardrails run inline on live agent traffic, which raises the cost of removing them once agents depend on them, but how its components are sold or renewed together is undisclosed, and the company is too new to show years of embedded integration. It holds no public certification in the reviewed record and names no anchor customers. The offensive research is difficult work and the founding team is strong, but the offensive-testing-plus-guardrail approach rests on techniques others can adopt, so neither the product nor the team is a barrier a funded rival cannot clear. The one asset that could compound is the attack data its red-team engine generates, which the record does not yet show as a named cross-customer asset.
| Dimension | Score | Rationale |
|---|---|---|
| Value Delivery Does the product sell software as the product, or judgment, trust, or accountability with software as the delivery mechanism. | 1/3 | General Analysis delivers software the customer configures and runs, a red-teaming platform, a guardrail SDK, and an MCP Guard proxy, with no managed service accepting accountability in the record. That is the software-product level. |
| Switching Cost How expensive leaving is for a customer: data portability, integrations, learned workflows, network effects, regulatory data residency. | 2/3 | Its runtime guardrails run inline on live agent traffic, so removing them once agents depend on them takes real effort, but the components are technically distinct with undisclosed commercial bundling, and the company is too young to show years of embedded integration. |
| Compliance Moat Whether certifications, liability acceptance, or audit trails block an easy replacement. | 1/3 | The reviewed record shows no public security attestation for General Analysis. A probe of its trust surfaces on July 3, 2026 found no SOC 2, ISO, or comparable certification, so compliance is not a barrier today. |
| Problem Complexity Whether the product requires ML, optimization, real-time systems, or years of specialized expertise. | 3/3 | Training adversarial agents that map a customer's tools and permissions and generate multi-step exploits is difficult offensive-research work that a generic engineering team cannot quickly reproduce. |
| Buyer Profile Whether buyers are SMB operators, mid-market IT teams, or regulated enterprises and governments with procurement gates. | 2/3 | General Analysis targets enterprise security teams responsible for production AI agents, but names no customer publicly, so its proven buyer relationships are still thin. |
| Layer Whether the product is an end-user application, a platform with application features, or infrastructure other applications depend on. | 2/3 | Part of the product runs inline on live agent traffic as runtime guardrails, but the red teaming and asset management do not, so the company does not yet own a chokepoint the customer's traffic must pass through. |
| Proprietary Data, Content, or IP Whether the product accumulates datasets, content licenses, or IP that a rival cannot recreate from scratch. | 1/3 | General Analysis trains its guards on its own red-team data and markets proprietary adversarial models, but that corpus is not named, independently benchmarked, or shown to compound across customers, so it is reproducible by a funded rival. |
General Analysis targets enterprise security teams responsible for autonomous AI agents the business puts into production. Its own materials name the buyers it wants to reach, spanning employee copilots, coding agents, customer-support agents, and copilots in healthcare, legal, and financial workflows.
The segment is defined by a shared problem rather than one vertical. Every named buyer runs agents that mix untrusted input with privileged access, so the company sells the same adversarial testing and runtime defense across industries rather than tailoring a separate product per sector.
The company trains adversarial agents that ingest a customer's runtime environment, map its full graph of tools and permissions, and generate multi-step exploits aimed at the seams between components. That offensive engine feeds the defensive side, where General Analysis ships guardrails as open checkpoints and as custom detectors trained on the same adversarial pipelines, plus an MCP Guard proxy that screens the tool output returned through MCP servers.
External visibility is notable this early. Security researcher Simon Willison cited the company's constructed attack scenario showing how a Supabase database connected to the Cursor coding agent could be drained through one poisoned support ticket, and independent press reported its red-team demonstration against live customer-service bots. No reproducible third-party benchmark of detection accuracy appears in the cited materials, so most capability evidence still comes from the vendor's own documentation.
General Analysis reaches buyers two ways. Its marketing site leads with enterprise demo and consultation requests, while its documentation offers a self-serve path: installing the command-line tool, logging in, and starting a 30-day free trial with no payment method. It names no customer publicly, and no named reference appears in the public record.
Its clearest go-to-market asset is reputation. The Y Combinator network, backing from Altos Ventures and Menlo Ventures, and coverage from Axios and Simon Willison's blog give a young company visibility, though that attention has not yet converted into public reference customers.
General Analysis does not post public pricing for its paid tiers. It offers a free Guardrail Basic trial through its command-line tool, but no rate card for paid usage appears on its public pages.
The unit it charges paying customers by is not disclosed, so it is not possible to tell from public materials whether the company prices by agents covered, calls inspected, engagements, or seats. For a buyer, that means cost discovery beyond the free trial depends on a sales conversation.
The product is delivered through a mix of customer-run and GA-managed components. The GA SDK installs guardrails, open guard checkpoints can be self-hosted or invoked through the platform, and MCP Guard runs as a local proxy that routes tool output to GA's guardrail endpoint.
Custom detectors are delivered through managed endpoints trained against a customer's policies and red-team data. The mix of self-serve SDK and managed custom guards means the company can start a customer on open tools and layer bespoke detection as policy needs grow.
The reviewed record shows no public security attestation, and probes of the trust and security paths found none when re-run on July 15, 2026. A probe of its homepage footer, its /security and /trust paths, and its trust and security subdomains on July 3, 2026 found no SOC 2, ISO, or comparable certification.
The absence is normal for a company at seed stage but matters here because the buyer is a security team granting an agent broad access. Until General Analysis can show its own audited controls, trust rests on its research reputation rather than on independent assurance.
General Analysis positions itself as a neutral layer across the AI stacks enterprises already run rather than a single-model tool. Its materials describe securing workplace copilots from several vendors, coding agents, and MCP servers, and it publishes open guard checkpoints that customers can self-host.
The ecosystem strategy is breadth over lock-in. By working across model providers, copilots, coding agents, and MCP servers and shipping some guards openly, the company aims to be adopted wherever agents run, though that same openness means the integration points are not themselves a barrier a rival could not match.
The founding team carries research and engineering backgrounds relevant to the problem. Co-founder Rez Havaei is quoted by Axios, and the company's own account describes a founding team that built reinforcement-learning infrastructure at DeepMind, trained models at Jane Street, NVIDIA, and Cohere, and published at NeurIPS, ICLR, and ICML.
The credential is real but early. Public sources do not name a prior exit or detail a sustained publication record, and the strongest outside recognition so far is Simon Willison's citation of the Supabase and Cursor attack scenario.
| Id | Source | Tier | Accessed |
|---|---|---|---|
| f1 | About General Analysis (what the company does) | official | 2026-07-03 |
| f2 | Business Wire seed announcement, About boilerplate. A Y Combinator directory capture of 2025-12-01 listed Founded 2024 for the batch entity | press | 2026-09-10 |
| f3 | General Analysis seed announcement (location) | official | 2026-07-03 |
| f4 | Tech Startups: General Analysis raises $10M seed (investors) | press | 2026-07-03 |
| Id | Source | Tier | Accessed |
|---|---|---|---|
| s1 | General Analysis homepage “General Analysis helps security teams adversarially test, monitor, and protect AI agents and systems in production.” | official | 2026-07-03 |
| s2 | About General Analysis (approach) “We train proprietary adversarial agents that digest a customer's runtime environment, map the full graph of tools and permissions, and generate multi-step exploits that target the seams between components.” | official | 2026-07-03 |
| s3 | General Analysis seed announcement (founding team) “Our founding team of top AI safety and security researchers has built RL infrastructure at DeepMind, trained state-of-the-art models at Jane Street, NVIDIA, and Cohere, and published leading research at top conferences, including NeurIPS, ICLR, ICML, and more.” | official | 2026-07-03 |
| s4 | General Analysis guardrails documentation “Our safety stack ships in two guard families, both trained on the same adversarial pipelines we run for production deployments. Public guards, Open checkpoints published on Hugging Face. Custom enterprise guards, Bespoke detectors we train against your policies and red-team data.” | official | 2026-07-03 |
| s5 | General Analysis MCP Guard documentation “You'll be automatically enrolled in a 30-day free trial for Guardrail Basic. No payment method required! The GA CLI and MCP Guard proxy are implemented in Python.” | official | 2026-07-03 |
| s6 | Axios: Agentic AI security startup General Analysis raises $10M “General Analysis, which helps companies safely deploy agentic AI, has raised $10 million in seed funding led by Altos Ventures, co-founder Rez Havaei tells Axios Pro. The company aims to hit $2 million in revenue in the next 12 to 18 months and then will look to raise a Series A.” | press | 2026-07-03 |
| s7 | Tech Startups: General Analysis red-team demonstration “In March, a single adversarial agent persuaded 50 live customer service bots to hand out more than $10 million in fake perks. Million-dollar gift cards. Years of free services. Anything it could extract. Each target took only a few minutes. Out of 55 systems tested, just five held the line.” | press | 2026-07-03 |
| s8 | Simon Willison: Supabase MCP can leak your entire SQL database “In this case, General Analysis identify all three components in a single MCP, the Supabase MCP. They imagine a scenario where a developer asks Cursor, running the Supabase MCP, to list the latest support tickets.” | press | 2026-07-03 |
| s9 | General Analysis raises $10M seed (location and investors) “General Analysis, a San Francisco startup building security infrastructure for agentic AI. The round included participation from 645 Ventures, Menlo Ventures, Y Combinator, and other strategic investors and angels.” | press | 2026-07-03 |
| s10 | General Analysis homepage trust probe 2026-07-03 (footer badges, /security and /trust paths, trust subdomains), no SOC 2 or ISO attestation found “Understand your AI risk.” | official | 2026-07-03 |
| Id | Source | Tier | Accessed |
|---|---|---|---|
| s1 | General Analysis homepage “General Analysis helps security teams adversarially test, monitor, and protect AI agents and systems in production.” | official | 2026-07-03 |
| s2 | About General Analysis (approach) “We train proprietary adversarial agents that digest a customer's runtime environment, map the full graph of tools and permissions, and generate multi-step exploits that target the seams between components.” | official | 2026-07-03 |
| s3 | General Analysis seed announcement (founding team) “Our founding team of top AI safety and security researchers has built RL infrastructure at DeepMind, trained state-of-the-art models at Jane Street, NVIDIA, and Cohere, and published leading research at top conferences, including NeurIPS, ICLR, ICML, and more.” | official | 2026-07-03 |
| s4 | General Analysis guardrails documentation “Our safety stack ships in two guard families, both trained on the same adversarial pipelines we run for production deployments. Public guards, Open checkpoints published on Hugging Face. Custom enterprise guards, Bespoke detectors we train against your policies and red-team data.” | official | 2026-07-03 |
| s5 | General Analysis MCP Guard documentation “You'll be automatically enrolled in a 30-day free trial for Guardrail Basic. No payment method required! The GA CLI and MCP Guard proxy are implemented in Python.” | official | 2026-07-03 |
| s6 | Axios: Agentic AI security startup General Analysis raises $10M “General Analysis, which helps companies safely deploy agentic AI, has raised $10 million in seed funding led by Altos Ventures, co-founder Rez Havaei tells Axios Pro. The company aims to hit $2 million in revenue in the next 12 to 18 months and then will look to raise a Series A.” | press | 2026-07-03 |
| s7 | Tech Startups: General Analysis red-team demonstration “In March, a single adversarial agent persuaded 50 live customer service bots to hand out more than $10 million in fake perks. Million-dollar gift cards. Years of free services. Anything it could extract. Each target took only a few minutes. Out of 55 systems tested, just five held the line.” | press | 2026-07-03 |
| s8 | Simon Willison: Supabase MCP can leak your entire SQL database “In this case, General Analysis identify all three components in a single MCP, the Supabase MCP. They imagine a scenario where a developer asks Cursor, running the Supabase MCP, to list the latest support tickets.” | press | 2026-07-03 |
| s9 | General Analysis raises $10M seed (location and investors) “General Analysis, a San Francisco startup building security infrastructure for agentic AI. The round included participation from 645 Ventures, Menlo Ventures, Y Combinator, and other strategic investors and angels.” | press | 2026-07-03 |
| s10 | General Analysis homepage trust probe, re-run 2026-07-15 (footer badges, /security and /trust paths, trust subdomains), no SOC 2 or ISO attestation found “Understand your AI risk.” | official | 2026-07-03 |
This site is an experimental research aid created by Zeltser Security Corp. All its data gathering and analysis was performed autonomously without human review, and it can contain errors of fact, interpretation, and judgment that a human reviewer might catch.
The analyses are statements of opinion, not statements of fact. Machine analysis produced the scores, summaries, and matrix placements by weighing the public sources each page cites, and reasonable people can weigh the same sources differently. Where a page states a fact, it cites the public source and the date it was checked, and the statement is only as accurate as that source. Unless a profile expressly says otherwise, the analysis involves no hands-on testing and no independent validation of any company's products or services.
Nothing here is professional, security, legal, financial, investment, or purchasing advice, and nothing here is a recommendation to invest in, do business with, or avoid any company. Inclusion of a company is not an endorsement, and absence of a company is not a judgment about it. Reading this site creates no advisory or client relationship. Verify any detail you plan to act on against the vendor's current materials.
The content is provided "as is" and "as available," with all warranties disclaimed, express or implied, including merchantability, fitness for a particular purpose, accuracy, and non-infringement. No entry is warranted to be complete, current, or correct. Companies change, vendors update their claims, sources can be wrong, and automated analysis can misread them.
To the fullest extent permitted by law, the operator, Zeltser Security Corp, is not liable for any damages that arise from using this site or relying on its content, including direct, indirect, incidental, special, and consequential damages and lost profits, even if advised that such damages were possible. If you are dissatisfied with the site or disagree with these terms, your remedy is to stop using it.
Entries link to vendor pages, press coverage, and other external sites that Zeltser Security Corp does not control and is not responsible for. A link is not an affiliation with the destination or an endorsement of it. Product and company names and trademarks are the property of their owners, used here nominatively to identify the companies described. Short quotations from cited sources appear for identification and commentary.
Use, quotation, automated retrieval, and redistribution of the content are governed by the Terms of Use at cybercompanyprofiles.com/terms, which permit personal and internal business use with attribution and prohibit republication and resale.