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.
Gray Swan AI sells Cygnal, software that blocks attacks on AI models and agents and their unsafe outputs, to enterprises and frontier AI labs. Its Shade agent attacks customers’ models to find flaws. Gray Swan trains both on attacks from its Arena, where over 15,000 researchers compete to break models. Carnegie Mellon professors founded the company in 2023, the year they published an automated way to make language models ignore safety rules. It raised a $40 million Series A co-led by Wing Venture Capital and Madrona. Its named customers include OpenAI, Anthropic, and Google DeepMind, and Forbes reports Snowflake uses its software to test Snowflake’s agents. Frontier labs, which Forbes reports supply most of its revenue, could sell their own runtime screening to Cygnal’s enterprise buyers.
| Description | Gray Swan's Cygnal is a runtime guardrail that sits inline between users, models, agents, and their tools, classifying adversarial inputs and unsafe outputs in real time for enterprise AI applications in production. | [f1] |
|---|---|---|
| Founded | 2023 | [f2] |
| HQ | Pittsburgh, Pennsylvania, USA | [f2] |
| Latest funding | Series A, USD 40M (May 2026), co-led by Wing Venture Capital and Madrona | [f2] |
| Deployment | Hybrid, SaaS, Self-hosted | [f3] |
| Compliance | Cyber Essentials, SOC 2 Type 2 | [f3] |
| Product | What it does |
|---|---|
| Gray Swan Cygnal | Gray Swan Cygnal: Inline runtime guardrail that screens prompts, model responses, and agent tool calls, blocking prompt injection, jailbreaks, and unsafe outputs against custom policies. |
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 Coding and Orchestration Tools AI coding tools and agentic orchestration tools on user devices, plus their plugins, skills, hooks, system prompts, scaffolding, harnesses, configuration settings, and MCP clients. | ||||||
| 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 Traffic Paths to external and self-hosted AI services, MCP tool channels, agent-to-agent calls, model-registry downloads, and egress to unapproved AI services. AI gateways, LLM routers, and MCP gateways steer traffic along those paths. The steering decisions and their inputs belong here too: routing policies, MCP server registries, and agent naming and discovery services. | ||||||
| AI Model Model weights, fine-tuning checkpoints, model cards, registries, an AI bill of materials (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. |
Gray Swan Cygnal is an inline runtime guardrail that screens prompts, model responses, and agent tool calls, blocking prompt injection, jailbreaks, and unsafe outputs against custom policies. It is mapped to the AI Defense Matrix. [f4]
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. | 4/5 | Gray Swan names two buyers, enterprises running AI agents and the labs building frontier models, and names the harm as prompt injection, jailbreaks, and unauthorized tool use. Gray Swan quantified it in a research paper with NIST’s CAISI and the UK AI Security Institute, and NIST’s own blog reports a successful hijacking of all 13 frontier models tested. [s3, s2, s18, s19] |
| Capability Depth How specific the technical capabilities are, with evidence beyond marketing claims such as docs and third-party validation. | 4/5 | Gray Swan’s docs specify Cygnal’s completions proxy, its required policy headers, the refusal it returns on a violation, and a separate monitoring API that scores messages from 0 to 1. TrueFoundry’s own documentation explains how to wire Cygnal into its AI gateway, and Forbes quotes a Snowflake research scientist saying Gray Swan can guard against very subtle attacks. [s10, s11, s3, s22, s16] |
| Market Timing Whether the market is ready for this product, with evidence that buyers are actively seeking solutions. | 3/5 | The enabler is credible and dated: AI agents that read outside content, which NIST’s CAISI describes as facing growing hijacking risk in its March 2026 research with Gray Swan. Buyer-side demand in the last year rests on one documented kind, customer adoption, which Forbes reports through Snowflake’s use and frontier-lab spending. The broader customer count comes from the company, so the record shows one kind of buyer signal rather than two. [s18, s16, s17] |
| Team Credibility Demonstrated domain expertise with public signals such as prior exits, publications, and industry recognition. | 5/5 | Forbes reports that founders Matt Fredrikson and Zico Kolter, Carnegie Mellon professors, discovered the 2023 attack dubbed the mother of all jailbreaks, and that Kolter chairs the OpenAI Foundation board’s safety and security committee. Forbes also counts Gray Swan in 11 frontier model system cards, and NIST’s CAISI took it on as a research partner, independent evidence of sustained leadership in the field. [s16, s24, s18, s2] |
| GTM Proof Evidence of actual traction (customers, revenue signals, partnerships) beyond stated intentions. | 5/5 | Forbes reports in its own voice that Gray Swan has 20 enterprise customers, names OpenAI as its first customer, and quotes a Snowflake research scientist saying Snowflake uses its software to test its agents. Earlier Forbes reporting describes contracts with OpenAI, Anthropic and the UK AI Safety Institute, and NIST’s CAISI names Gray Swan as a research partner. That is customer scale reported by third parties plus public customer testimony. [s16, s15, s18] |
| Funding Efficiency Whether funding matches go-to-market ambition, with signs of capital-efficient growth. | 3/5 | Gray Swan raised a $40 million Series A in May 2026 at a valuation Forbes puts at $200 million, after early rounds Technical.ly, citing PitchBook, puts at $5 million in each of 2024 and 2025. The raise is proportional to a company shipping Cygnal, Shade, the Arena, and red-teaming services, and the cited record gives no revenue figure to test output per dollar. [s16, s17, s3, s4, s5] |
| Category Clarity Whether the company creates or fits a recognizable category that buyers can quickly place in their stack. | 4/5 | Outside sources place Gray Swan without its own terminology. Forbes calls it a security provider to frontier labs, Technical.ly calls it an AI security startup that safety-tests models and agents, and NIST’s CAISI names it a partner in AI agent red-teaming research. That places Gray Swan in the clearly emerging AI security and red-teaming category, multiply evidenced, though its three-part platform naming is its own. [s16, s17, s18] |
| Incumbent Defensibility How vulnerable the core value proposition is to absorption as a feature by a platform vendor. | 4/5 | Forbes reports that Gray Swan trains its Shade and Cygnal agents on data from the human red teamers in its Arena, some 15,000 security professionals, and calls that human data its edge. Gray Swan’s own Shade page says every vulnerability Shade finds can become a Cygnal detection rule. A running supply of fresh human attack data is a data flywheel that bundling comparable features alone would not recreate, even though Gray Swan shares model-specific competition data with frontier labs. [s16, s4, s19] |
Gray Swan AI sells protection and testing for AI models and agents to two buyers, enterprises putting AI into production and the frontier labs building the models. Its about page describes automated adversarial testing, continuous red teaming, and runtime AI security for both groups. The Cygnal page also covers regulated and high-stakes deployments, where a missed classification can mean a breach or a violation.
The harm Gray Swan targets is an AI agent that someone manipulates into saying or doing the wrong thing. Its Cygnal page lists prompt injection, including injections delivered through retrieved content and tool output, along with jailbreaks, unsafe outputs, and unauthorized tool use. Gray Swan says those failures extend past bad answers to an agent that calls an API it shouldn’t or runs a workflow it was manipulated into.
Gray Swan measured the problem in research it published with US and UK government institutes. A NIST research blog from March 2026 says CAISI partnered with Gray Swan, the UK AI Security Institute, and frontier labs on a paper about a Gray Swan competition, whose entrants hijacked every one of 13 frontier models at least once. The paper reports attack success rates from 0.5% to 8.5% across the models. [s3, s2, s18, s19]
Gray Swan Cygnal is a runtime guardrail that screens traffic between users and a model, an agent and its tools, and a retrieval pipeline and its content. The product page says Cygnal monitors every prompt, response, and tool call and lets a team set its own thresholds and actions, from blocking to flagging or rewriting. It deploys as SaaS, on-premises, or in a customer’s VPC.
The public docs describe how Cygnal works in practice. The completions proxy requires an explicit policy, and when Cygnal detects a violation it cuts the model’s response, returns a refusal, and marks the finish reason as a violation. A separate monitoring API scores messages for policy violations from 0 to 1, and Gray Swan’s TrueFoundry announcement describes an observe-only mode that logs violations before a team enforces them.
Gray Swan Shade is the testing counterpart, an agent that runs adaptive attack campaigns against a customer’s model, guardrails, and deployment. The Shade page says the attacks come from techniques found in the Arena and that findings include reproductions and severity ratings. Gray Swan also says any vulnerability Shade finds can become a Cygnal detection rule, which ties testing to protection. [s3, s10, s11, s4, s8]
Gray Swan positions Cygnal against guardrails trained on older attacks. The Cygnal page says traditional guardrails learn from attacks that stopped working months ago, while Cygnal trains on attacks that work now, sourced from more than 15,000 red teamers. Forbes reported the same point in its own words, that the Arena’s human data is Gray Swan’s edge against increasingly capable models.
That attack data is also shared with frontier labs and the US and UK AI safety institutes. Arena entrants license their submissions to Gray Swan and its partners. Gray Swan’s competition paper says it shares model-specific attack data with the frontier labs and the full dataset with the UK and US AI safety institutes. Gray Swan shares model-specific competition data with the respective frontier labs.
The frontier labs supply most of Gray Swan’s revenue and could also compete with it. Forbes reports that frontier labs make up a majority of its revenue, and any of those labs could add runtime screening to the models and agent platforms it sells. [s3, s16, s19, s14]
Gray Swan names its customers, and an independent outlet confirms two of the relationships. Its Trust Center lists Meta, ByteDance, xAI, Deloitte, Anthropic, Google DeepMind, Amazon, and OpenAI under the heading “Gray Swan is trusted by.” Forbes reports that OpenAI became its first customer less than a month after launch, and that Snowflake uses Gray Swan’s software to test its Cortex Code and Snowflake Intelligence agents.
Gray Swan reports about 20 customers, and most of its revenue comes from the labs. Gray Swan’s Series A announcement says it works with over 20 customers across frontier labs and global enterprises. Forbes reports 20 enterprise customers and says frontier labs make up a majority of revenue. No revenue figure appears in the cited sources.
Gray Swan uses integrations to reach enterprises that already run AI platforms. The Series A announcement describes a native integration with Snowflake’s AI ecosystem, and Snowflake Ventures joined the round. In July 2026 Gray Swan announced a Cygnal integration with TrueFoundry’s AI gateway, which TrueFoundry documents in its own integration guide. [s12, s16, s7, s8, s17]
Gray Swan’s founders are the researchers behind an automated 2023 jailbreak. Forbes reports that Carnegie Mellon professors Matt Fredrikson, the CEO, and Zico Kolter, the chief scientist, founded the company in 2023. They had discovered an attack dubbed the mother of all jailbreaks. Gray Swan’s research page describes it as an automated jailbreaking method published in July 2023.
The founders work with the organizations that evaluate frontier models. Forbes reports that Kolter chairs the safety and security committee on the OpenAI Foundation’s board and that Gray Swan is cited in 11 frontier model system cards. NIST’s CAISI partnered with Gray Swan and the UK AI Security Institute in 2026 to analyze data from a Gray Swan competition.
Gray Swan has added commercial operators to its research-led leadership. Gray Swan’s about page lists Rob Jenks as chief strategy officer and Spencer Whitman as chief product officer. Jenks joined in January 2026 from Tanium, where Gray Swan says he served as senior vice president of strategy and innovation. [s16, s24, s18, s2, s9, s15]
Gray Swan publishes a SafeBase Trust Center that lists SOC 2 Type 2 and Cyber Essentials under Compliance. Procurement teams can request access to Gray Swan’s security documentation.
Gray Swan offers deployment choices for buyers who want control over where an inline product runs. Gray Swan offers Cygnal as SaaS, on-premises, or in a customer’s VPC, so a buyer can choose where the screening runs. The TrueFoundry integration offers a fail-open mode, in which a Cygnal outage leaves the application running without screening. [s12, s3, s8]
| Company | Relationship | Note | Compare |
|---|---|---|---|
| Lakera | competes with | Competes for the same enterprise buyer of runtime guardrails that screen prompts and responses, the market Cygnal sells into. | |
| Lasso Security | competes with | Competes for enterprise buyers of runtime protection for model and agent traffic, the same purchase Cygnal targets. | |
| Prompt Security | competes with | Competes for the enterprise runtime AI security budget that Cygnal addresses. | |
| TrojAI | competes with | Competes on both halves of Gray Swan’s offer, adversarial testing and runtime defense, for the same enterprise buyer. | 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. |
| Adversa AI | adjacent | Adjacent through AI red teaming, the motion Shade and Gray Swan’s private red-teaming services address. | |
| OpenAI | adjacent | Forbes reports OpenAI as Gray Swan’s first customer, and as a model and agent platform it could offer its own runtime screening. |
Add analyzed competitors to compare them side by side with Gray Swan AI.
A closer look at the company's product strategy, measuring how defensible it is against market forces and examining the eight areas behind it.
press the advantage
Gray Swan trains Cygnal, which screens AI model and agent traffic, and its Shade attack agent on attacks from its Arena, a global competition with over 4 million submitted attempts. A rival would need its own supply of current attacks and years of adversarial machine-learning expertise to match those products. Entrants license their submissions to Gray Swan and to its partners, so Gray Swan does not hold that data exclusively. A customer that leaves Cygnal would have to rebuild the thresholds, actions, and routing rules it configured there. Frontier labs, which Forbes reports supply most of Gray Swan’s revenue, could build their own screening. Snowflake and TrueFoundry, whose platforms integrate Cygnal, could add screening and spare their customers a separate Cygnal purchase.
| Dimension | Score | Rationale |
|---|---|---|
| Value Delivery Does the product sell software as the product, or judgment, trust, or accountability with software as the delivery mechanism. | 2/3 | Beside the Cygnal and Shade software, Gray Swan sells private red-teaming engagements in which hand-selected Arena experts test a customer’s AI and deliver severity classifications and remediation roadmaps. Its offer to frontier labs pairs a custom automated attacker with that expert work and custom research, so code and expertise blend. |
| Switching Cost How expensive leaving is for a customer: data portability, integrations, learned workflows, network effects, regulatory data residency. | 2/3 | A customer that leaves Cygnal has to rebuild the custom policies, thresholds, and routing it configured, which is real friction. The cited record describes adoption as a two-line code change or a gateway setting and documents no mechanism that makes leaving costlier, so the migration effort is not sized. |
| Compliance Moat Whether certifications, liability acceptance, or audit trails block an easy replacement. | 1/3 | Gray Swan’s Trust Center lists SOC 2 Type 2 and Cyber Essentials, attestations a funded rival can obtain through ordinary preparation. The cited sources tie no regulatory mandate or authorization to the product. |
| Problem Complexity Whether the product requires ML, optimization, real-time systems, or years of specialized expertise. | 3/3 | Cygnal classifies adversarial inputs, unsafe outputs, and tool calls at production latency, using models Gray Swan trains on Arena attack data. The team published an automated jailbreaking method and posts code on GitHub for a circuit-breakers defense described in an arXiv paper, machine-learning work that takes years of adversarial-AI expertise. |
| Buyer Profile Whether buyers are SMB operators, mid-market IT teams, or regulated enterprises and governments with procurement gates. | 3/3 | Gray Swan sells to frontier labs, large enterprises, and a government institute. Its Trust Center lists Amazon, Deloitte, and the major labs, Forbes reports Snowflake using its software, and Forbes reported partnerships and contracts with OpenAI, Anthropic, and the UK AI Safety Institute. |
| Layer Whether the product is an end-user application, a platform with application features, or infrastructure other applications depend on. | 2/3 | Cygnal is middleware that applications route model and tool traffic through, as a proxy or a gateway plug-in with its own policies and dashboard. That places it between an application and its model as a configurable product layer. |
| Proprietary Data, Content, or IP Whether the product accumulates datasets, content licenses, or IP that a rival cannot recreate from scratch. | 2/3 | Gray Swan retains over 4 million Arena attack submissions from outside red teamers under an irrevocable license entrants grant, and Forbes reports that the data trains Shade and Cygnal. The license also covers its partners, and Gray Swan shares competition data with labs and safety institutes, so the mostly non-public corpus is replicable with time and effort. |
Gray Swan AI sells to two segments, frontier AI labs and enterprises deploying AI. Its about page describes adversarial testing, red teaming, and runtime AI security for both. Its Cygnal page addresses enterprises with agents in production and regulated, high-stakes deployments.
The labs are the larger segment today. Forbes reports that frontier labs make up a majority of Gray Swan’s revenue and describes it as a security provider to OpenAI, Anthropic, Google DeepMind, Meta, xAI, and ByteDance. Gray Swan’s Trust Center lists those labs alongside Amazon and Deloitte under the heading “Gray Swan is trusted by.”
Enterprises are the segment the new funding targets. Forbes reports 20 enterprise customers and says the Series A money will help Gray Swan sell to more such businesses. It names Snowflake as one that uses Gray Swan’s software to test its agents.
Gray Swan Cygnal screens prompts, responses, and tool calls at runtime against policies each customer defines. Its docs describe a completions proxy that requires an explicit policy and returns a refusal when a policy is breached. A monitoring API scores messages from 0 to 1. The product page lets a team choose to block, flag, or rewrite.
Gray Swan Shade tests AI systems the way an attacker would. The Shade page describes an agent that adapts, escalates, and chains attack techniques against a customer’s model, guardrails, and deployment, and returns findings with reproductions and severity ratings.
Both products train on attack data from the Arena, Gray Swan’s red-teaming competition. Gray Swan says the Arena yields more than one million attack trajectories used to train the models behind Cygnal and Shade, and Forbes reported the same training use. That link lets the company update its classifiers as competitors find new attacks.
Gray Swan sells to labs and enterprises and offers platform integrations. Forbes reports OpenAI as its first customer and Snowflake as an enterprise that uses its software. Its Trust Center lists the labs, Amazon, and Deloitte among the organizations that trust it.
Two integrations put Cygnal inside platforms enterprises already run. The Series A announcement describes a native integration with Snowflake’s AI ecosystem. In July 2026 Gray Swan announced a Cygnal integration with TrueFoundry’s AI gateway, which TrueFoundry documents in its own guide. Gray Swan’s TrueFoundry announcement says customers turn Cygnal on through a gateway configuration change.
Gray Swan also sells services staffed from the Arena. Gray Swan’s red-teaming page sells private engagements staffed by hand-selected Arena performers. The Arena’s about page says top performers join a private network for paid work with AI labs and enterprise partners.
Gray Swan publishes a free tier for Cygnal and the cited pages give no paid list price. The docs set the free tier at 200,000 tokens per minute and 10,000,000 tokens per day.
The free tier’s quota behavior matters to anyone evaluating on it. The docs say that when a free-tier quota is reached, Cygnal still proxies the request to the model provider, just without filtering, so screening stops while traffic continues. The docs state this only for the free tier.
Cygnal runs as a hosted service, on-premises, or in a customer’s VPC. The product page offers SaaS, on-premises, and VPC deployments and says integration takes a two-line code change. The docs describe requests routed through a Cygnal proxy with an explicit policy on each call.
Teams can watch before they enforce. The monitoring API returns violation scores from 0 to 1 for review. Gray Swan’s TrueFoundry announcement describes an observe-only mode that logs violations so a team can tune thresholds before it enforces them.
Gray Swan keeps the classifiers current from its own research. The Cygnal page says the team that breaks current models continuously updates Cygnal, and the TrueFoundry integration offers a fail-open mode so a Cygnal disruption need not stop the customer’s application.
Gray Swan backs procurement reviews with two attestations. Its SafeBase Trust Center lists SOC 2 Type 2 and Cyber Essentials, with security documentation available on request.
Buyers also get deployment choice, and Gray Swan has outside recognition through system-card citations and a NIST research collaboration. A buyer can run Cygnal on-premises or in its own VPC. Forbes reports Gray Swan cited in 11 frontier model system cards, and NIST’s CAISI worked with Gray Swan on its competition data.
Gray Swan combines three parts, Cygnal for runtime protection, Shade for automated testing, and the Arena as a red-teaming community whose attack data feeds both. Gray Swan’s Series A announcement describes the three together as security coverage across the AI lifecycle.
The Arena runs as a community with its own rules and sponsors. Its homepage lists OpenAI, Anthropic, Google DeepMind, Meta, Amazon, and the AI Security Institute as sponsors. Its about page counts more than 4 million attack attempts and says entrants grant Gray Swan and its partners an irrevocable license to use and share their submissions. Gray Swan’s competition paper open-sources the competition environment and a small set of attacks, and shares fuller data with labs and government safety institutes.
Outward, Gray Swan plugs Cygnal into other vendors’ platforms. The Snowflake and TrueFoundry integrations put Cygnal inside an AI data platform and an AI gateway, so enterprises can add screening without adopting a separate Gray Swan interface.
Gray Swan’s founders built their reputation breaking AI models. Forbes reports that Carnegie Mellon professors Matt Fredrikson and Zico Kolter founded the company in 2023 after discovering the attack dubbed the mother of all jailbreaks. Gray Swan’s research page describes it as an automated jailbreaking method published in July 2023.
Kolter and Gray Swan both take part in frontier-model safety evaluation. Forbes reports that Kolter chairs the OpenAI Foundation board’s safety and security committee, and NIST’s CAISI partnered with Gray Swan and the UK AI Security Institute on competition research.
Rob Jenks brought enterprise operating experience from Tanium. Rob Jenks joined as chief strategy officer in January 2026 from Tanium, and the about page lists Spencer Whitman as chief product officer.
| Id | Source | Tier | Accessed |
|---|---|---|---|
| f1 | Gray Swan: Cygnal product page | official | 2026-09-22 |
| f2 | Technical.ly: Gray Swan raises $40M Series A | press | 2026-09-22 |
| f3 | AI Defense Matrix Catalog entry | other | 2026-06-13 |
| f4 | AI Defense Matrix Catalog mapping | other | 2026-09-22 |
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.