Ethical Proxy Use Starts With Data Boundaries

Ethical proxy use isn't about whether scraping is legal. It's about whether your data practices are defensible at every stage of the pipeline.

The proxy industry spent years arguing about legality. That debate is largely settled. Scraping publicly available data isn't a federal crime in the US, and most of Europe has followed suit. But legality was never the whole question. The more important question, the one courts, regulators, and enterprise legal teams are now asking, is whether your data practices hold up at every point in the chain: what you collect, how you collect it, where your infrastructure comes from, and what you do with the data after.

⏱ 13 min read       Updated July 2026

Key takeaways

• Scraping public data is legal in the US and most of Europe. But legal and ethical are not the same standard, and the gap between them is where most compliance failures happen
• The Four Data Boundaries framework defines what ethical proxy use actually looks like: what you collect, how you collect it, where your infrastructure comes from, and what you do with the data after
• Public data is not automatically permissible data. Personal information visible on a public page still requires a documented lawful basis under GDPR before you collect it
• From August 2, 2026, robots.txt and ai.txt opt-outs carry legal weight under the EU AI Act. Ignoring them is no longer a convention violation, it's a compliance risk
• In 2026 alone, the FBI and Google disrupted two major residential proxy networks built on millions of hijacked consumer devices. Buyers using those networks had no idea; data provenance includes infrastructure provenance
• An Acceptable Use Policy with no enforcement mechanism is a liability shield, not an ethics framework. The difference matters when enterprise procurement teams start asking for proof
• Proxy providers with a verifiable track record spanning years of clean operation offer something no 2025 entrant can replicate, and in a market defined by rebrands and resellers, that track record is auditable in a way no feature page is
• The EU AI Act, new US state privacy laws, and a wave of AI training data lawsuits are all moving in the same direction: more documentation, more disclosure, more accountability. The buyers and operators who build these practices now will navigate that environment differently from those who don't.

The proxy is infrastructure. The ethics live in the decisions surrounding it.

This article introduces a framework, four data boundaries, that defines what ethical proxy use actually looks like in practice. It speaks to both sides of the market: buyers deploying proxies at scale, and operators building the infrastructure those buyers depend on. These aren't laws. They're the layer above the law: the practices that keep you defensible when regulation catches up, and in 2026, it's catching up fast.

The Legal Question Is Settled. The Ethical Question Isn't.

Before introducing the framework, it's worth understanding exactly where the law stands, because clarity here reveals how much space exists between "legal" and "ethical." Three landmark cases define the current landscape.

hiQ v. LinkedIn (Ninth Circuit, 2022) established that scraping publicly available data is not "unauthorized access" under the Computer Fraud and Abuse Act (CFAA). The court held that publicly accessible web pages cannot be protected by a "without authorization" claim, meaning scraping public data is not a federal crime. This precedent stands and is followed in most US courts.

Meta v. Bright Data (January 2024) reinforced this further. A federal judge ruled that Meta's Facebook and Instagram Terms of Service only prohibit logged-in scraping, not logged-off scraping of publicly accessible content. The court applied traditional contract interpretation to hold that Bright Data did not "use" Facebook when it scraped public pages without being authenticated. Meta dropped the lawsuit in February 2024.

Van Buren v. United States (Supreme Court, 2021) tightened the CFAA's scope from the other direction: the "exceeds authorized access" clause applies only to information you have no right to access at all, not to data you access legitimately but use for a purpose the site owner dislikes.

So far so good for scrapers. But here's where hiQ's story turns instructive: hiQ won on CFAA and lost on contract. Because hiQ had accepted LinkedIn's Terms of Service (which prohibit scraping), that contractual agreement was enforceable, even though the underlying scraping wasn't a federal crime. The lesson is that legality and contractual exposure are two separate risk surfaces, and winning one doesn't settle the other.

More telling is what happened in France in 2025. The data protection authority CNIL fined KASPR €240,000 for collecting LinkedIn data without appropriate consent. The collection was arguably legal under CFAA analysis. It was non-compliant under GDPR because GDPR applies to whose data is collected, not where the collector is based or what US law permits.

This is the gap the four data boundaries are designed to address. Legality is the floor. Ethics is the ceiling you choose to build. Between them is where most of the real risk lives.

The Four Data Boundaries of Ethical Proxy Use

Ethical proxy use lives across four distinct decisions made before, during, and after any data collection operation. Getting all four right is what makes a proxy operation defensible. Getting any one wrong is where exposure begins.

Boundary 1: What Data You Collect

The clearest legal and ethical line in proxy-assisted data collection is also the simplest: personally identifiable information (PII) is a fundamentally different category from non-personal signals, and treating them the same way is where most compliance failures begin.

Publicly available product prices, search engine results, business listings, news headlines, stock data, sports scores — these are defensible fields. They're signals about the world, not about individuals. Scraping them with proxies at scale, for business intelligence, price monitoring, or research, sits inside a well-established legal and ethical space.

Names, email addresses, phone numbers, profile photographs, location history, and behavioral data are a different matter, even when they're technically visible on a public page. Under GDPR Article 6, collecting personal data requires a lawful basis. For commercial scraping operations, "legitimate interest" is the most commonly applied basis, but it requires documented justification that your interest outweighs individual privacy rights. Regulators expect that documentation to exist before collection begins, not assembled retrospectively when an audit arrives.

The "public equals permissible" assumption is the most common first-boundary failure. A photograph on a public social media profile does not grant permission to scrape it, store it, and build a commercial product with it. The European Data Protection Board's Opinion 28/2024 tightened this specifically for AI: it endorsed legitimate interest as a possible basis for scraping personal data to train AI models, but only under a strict three-step proportionality test. Facial images scraped without targeted consent are explicitly banned under the EU AI Act, which entered full enforcement on August 2, 2026.

The practical discipline here is simple but frequently skipped: before any scraping operation, map each data field to a specific purpose and a documented legal basis. If you cannot articulate why you need a field and what right you have to it, don't collect it. Data minimization is not just a GDPR principle. It's the most reliable way to keep your operation inside defensible boundaries as regulations tighten.

Boundary 2: How You Collect It

The second boundary governs method. You can have a lawful right to collect data and still create significant legal and ethical exposure through the way you collect it. The most significant shift in collection-method ethics in 2026 involves robots.txt. This file, which websites use to communicate crawling preferences to automated agents, was previously treated as a convention, something well-behaved bots respected and bad actors ignored. In 2026, it carries legal weight.

Under the EU Copyright Directive's Article 4 text-and-data-mining exception, website operators can opt out of having their content used for data mining by using machine-readable signals, including robots.txt directives and the emerging ai.txt standard. From August 2, 2026, EU AI Act Article 53(1)(c) requires all general-purpose AI model providers to have a documented policy for complying with this copyright opt-out mechanism. A website's robots.txt is now a legally meaningful statement of intent for AI training data collection. Ignoring it is no longer a violation of convention. It's a compliance risk with fines up to €15 million or 3% of global turnover for GPAI providers under the AI Act.

For proxy-based collection more broadly, the ethical method standards are:

• Rate limiting is not just courtesy. Aggressive request rates that degrade a target server's performance create liability exposure under computer misuse laws in multiple jurisdictions. The ethical standard is collecting at a rate the target can sustain without service degradation, regardless of whether you could technically send more.
• Authentication barriers draw the clearest line. Bypassing a login requirement, a paywall, or any technical access control is a criminal-access risk in every jurisdiction that has computer misuse legislation. The hiQ and Meta v. Bright Data precedents protect logged-off scraping of public content. They do not protect circumventing systems designed to restrict access.
• Login scraping is contractually constrained even when the data behind the login is publicly visible elsewhere. If you have accepted a site's Terms of Service by creating an account, logging in, or clicking through an agreement, those terms are contractually enforceable against you regardless of whether the underlying collection would otherwise be legal. This is precisely where hiQ failed, and where LinkedIn won its case on contract grounds after losing on CFAA.

The ethical method standard, taken together: collect the minimum data you need, at a rate the target can sustain, from surfaces you have a legitimate right to access, through infrastructure you can defend.

Boundary 3: Where Your Infrastructure Comes From

The third boundary is the one the proxy industry is most reluctant to discuss openly, and the one where the most significant harm has occurred.

The ethics of data collection extend to the infrastructure layer. Not just what you collect but through what. And in 2026, the infrastructure layer has been at the center of two of the most significant enforcement actions in the industry's history.

In January 2026, Google's Threat Intelligence Group, the FBI, Lumen Technologies, and the Shadowserver Foundation jointly disrupted IPIDEA, at the time one of the world's largest residential proxy networks. In July 2026, the same coalition disrupted NetNut, operated by Alarum Technologies, a publicly traded company on Nasdaq. At the core of NetNut's residential IP pool was the Popa botnet: over two million compromised consumer devices (smart TVs, streaming boxes, Android devices) recruited without meaningful user consent through deceptive SDKs embedded in unofficial apps and inexpensive off-brand hardware.

Both networks had extensive reseller ecosystems. Dozens of proxy brands were white-labeling this infrastructure without disclosing it and in many cases without knowing it. Buyers routing traffic through those networks were making requests through hijacked household devices without those households' knowledge. The device owners' bandwidth was being sold; their IP addresses were being used to serve requests they never agreed to; their devices were part of infrastructure generating revenue they never saw.

Using that infrastructure is not ethically neutral even for buyers who had no knowledge of the sourcing. The harm is real, and the chain is traceable. And as data provenance requirements expand, particularly under the EU AI Act, the question of what infrastructure your data was collected through will increasingly appear in compliance documentation.

For proxy operators, the third boundary requires documented, verifiable consent from every device in the network. The standard is clear: SDK transparency about what network participation means, opt-in compensation, clear and readable terms, and the ability to produce that documentation when enterprise buyers or regulators ask. An IP pool that cannot be traced to a sourcing mechanism the device owner knowingly agreed to is not ethically defensible regardless of what it costs per GB or how many IPs it contains.

For proxy buyers, the third boundary requires due diligence on sourcing before purchase. Ask your provider directly: how does a device join your residential network? What consent mechanism exists? Can you produce a sourcing policy document? Run a sample of IPs through a public ASN lookup, if the registered organization doesn't match the provider selling to you, you're in a reseller chain, and the question of what's at the other end of that chain deserves an answer.

For datacenter proxies, the sourcing question has a structurally simpler answer: the IPs come from commercial hosting infrastructure you can verify independently by ASN. The ethical questions shift to who else shares your subnets and what they use them for. Because subnet reputation is collective. A subnet with a history of abuse carries that history into every request made from it, including yours.

The insight that ties this boundary to the others: data provenance includes infrastructure provenance. You cannot claim ethical data collection practices while routing those requests through infrastructure built on device conscription.

Boundary 4: What You Do With the Data After

The fourth boundary is where ethical proxy use most frequently gets incomplete treatment because most compliance thinking stops at collection and doesn't extend to what happens next.

Collection and use are separate legal obligations under GDPR. Lawfully collecting data does not automatically authorize every use of it. The purpose you documented under Article 6 at collection time is the purpose you're authorized for. Using data collected for competitive price monitoring to build a commercial consumer product involving personal profiles is a purpose-limitation violation, even if the original collection was lawful.

Data minimization continues past collection. Fields that were relevant to your stated purpose at the time of collection may become irrelevant as projects evolve. GDPR requires deletion when data is no longer necessary for the purpose it was collected for. Building deletion workflows into your data pipeline from the beginning, rather than retrofitting them when a Data Subject Access Request arrives, is the difference between an operational burden and a compliance capability.

DSARs themselves deserve specific attention. If you've scraped personal data, even from public sources, and an individual requests deletion or access, GDPR requires a response within 30 days regardless of whether you're the original publisher of that data. The obligation attaches to data controllers, not just to platforms. "We scraped it from a public page" is not an exemption.

AI training data has made the fourth boundary urgent in a new way. The EU AI Act requires general-purpose AI model providers to publish a training-data summary documenting sources, methods, and compliance with copyright opt-outs, using a mandatory Commission template. This requirement creates a retrospective documentation problem for anyone who collected training data without provenance tracking: the data may have been lawfully collected at the time, but the inability to document its provenance today creates compliance exposure under tomorrow's standard.

The practical discipline for the fourth boundary mirrors the first three: build the documentation habit before you need it. Provenance records, field-level purpose mapping, and deletion workflows are significantly easier to establish before a collection operation scales than after it has generated terabytes of undocumented data. The specific questions to ask yourself and be able to answer are covered in the buyer defensibility framework below.

For Proxy Providers: Your Acceptable Use Policy (AUP) Is Your Ethics Infrastructure

An Acceptable Use Policy is not a legal formality. In 2026, it's the primary mechanism by which ethics travels down the proxy supply chain: from operator to buyer, and from buyer to the targets and individuals their operations touch.

The first principle: ethical providers vet not just the source of their IPs but the intent of their customers. If a company intends to use a proxy network for credential stuffing, DDoS attacks, fraud, or harvesting PII without a lawful basis, the provider's infrastructure becomes an enabler of that harm regardless of whether the AUP technically prohibits it on paper. A list of prohibited uses with no enforcement mechanism is a liability shield, not an ethical framework.

An operationalized AUP in 2026 has four components:

• Know Your Customer. Verify buyer identity and stated use case before granting access. This isn't bureaucracy. It's the first line of defense against your infrastructure being used for purposes that will eventually come back to the provider. Enterprise buyers increasingly expect it as part of their own vendor due diligence.
• Explicit use case restrictions. Prohibit credential stuffing, DDoS facilitation, PII harvesting without documented lawful basis, scraping behind authentication, and any use involving bypassing technical access controls. Make these specific enough to enforce, not broad enough to be meaningless.
• Active monitoring and enforcement. Review traffic patterns for signals of prohibited activity. Request rate anomalies, unusual target clustering, behavioral patterns consistent with account takeover or fraud. A proxy provider that has never terminated a customer for AUP violations is probably not monitoring for them.
• Supply chain documentation. Publish how IPs are sourced, who upstream partners are, and what consent mechanisms exist. This has shifted from a differentiator to a procurement requirement: enterprise legal and compliance teams in 2026 are asking for it before signing contracts, and providers who cannot produce it are losing deals to those who can.

There is also a longer-term argument here that goes beyond compliance. A provider that has operated under consistent ethical standards for fifteen years (through the emergence of GDPR, through the botnet era, through the AI data rush, through two major enforcement disruptions in a single year) has built something that cannot be replicated by a 2026 entrant: a verifiable track record. Every year of clean operation is a data point that no marketing claim can substitute for. In a market where buyers are increasingly asking hard questions about sourcing and history, that track record is an asset with compounding value.

For Proxy Buyers: The Defensibility Test

For buyers, the four data boundaries resolve into a single practical question: if you had to explain your proxy operation to a regulator, an enterprise procurement team, or a court, could you do it?

The defensibility test is five questions:

1. What data did you collect? Can you list each field and the lawful basis you documented for it?
2. What method did you use? Did you respect robots.txt and ai.txt opt-outs? Did you rate-limit appropriately? Did you stay outside authentication barriers?
3. Where did your infrastructure come from? Can your proxy provider produce a sourcing policy? Did you verify their ASN against their brand? Do you know who else shares your subnets?
4. What did you do with the data? Is it being used for the purpose you documented at collection? Do you have a retention schedule? Can you respond to DSARs?
5. How long have you kept it? Is data that has served its purpose being deleted, or accumulating indefinitely?

Operations that can answer all five are inside defensible data boundaries. Operations that cannot answer one or more carry risk proportional to their scale, and in 2026, the scale of enforcement is increasing alongside the scale of collection.

Where This Is Heading

The regulatory direction is unambiguous. The EU AI Act's enforcement teeth arrived on August 2, 2026: GPAI fines up to €15 million or 3% of global turnover, with mandatory training-data disclosure and copyright opt-out compliance. New US state privacy laws (Indiana, Kentucky, Rhode Island) became effective in 2026, extending the patchwork toward national coverage. LinkedIn sued Proxycurl in 2026, opening another front in ToS-based enforcement of scraping restrictions.

The AI training data lawsuits are creating a body of case law that will take years to settle but is moving consistently in one direction: toward more documentation, more disclosure, and more accountability across the collection chain. The cases involving the New York Times, Reddit, and class actions from content creators are not just about AI companies. They're about the infrastructure and practices that feed them data.

For proxy operators and buyers who have already built their practices around the four data boundaries, this trajectory represents confirmation rather than disruption. For those treating compliance as a future cost center, the enforcement timeline is becoming concrete.

The proxy industry has a choice about how it enters that environment. It can wait for regulation to force standards it should have adopted earlier. Or it can demonstrate that the infrastructure powering the web's data economy is capable of holding itself to a standard that earns trust rather than requiring it.

Final Thought

The four data boundaries: what you collect, how you collect it, where your infrastructure comes from, and what you do with it after, are not a compliance checklist. They're a way of thinking about data collection that makes the ethics visible before they become a legal problem.

The proxy market has seen two major botnet disruptions in one year, a wave of AI data lawsuits, and a major regulatory enforcement deadline, all in the same twelve months. The industry that emerges from this period will look different from the one that entered it. The operators and buyers who have built defensible practices will navigate the transition differently from those who haven't.

Ethical proxy use starts with data boundaries. It always has. The difference in 2026 is that the consequences of ignoring them are no longer hypothetical.

Frequently Asked Questions

Is web scraping with proxies legal in 2026?
Scraping publicly available data with proxies is generally legal in the United States and most of Europe. The Ninth Circuit's hiQ v. LinkedIn ruling (2022) established that public data scraping is not unauthorized access under the CFAA, and Meta v. Bright Data (2024) confirmed that logged-off scraping of publicly accessible content does not violate a platform's Terms of Service. However, scraping personal data triggers GDPR obligations regardless of where the collector is based, ToS violations create separate contract risk for logged-in users, and bypassing technical access controls is a criminal risk in every jurisdiction.

Do I need to respect robots.txt when scraping with proxies?
From August 2, 2026, yes for AI training data collection targeting the EU market. Under EU Copyright Directive Article 4, website operators can use machine-readable signals like robots.txt and ai.txt to opt out of having their content used for text and data mining. EU AI Act Article 53(1)(c) requires general-purpose AI model providers to have a documented policy respecting these opt-outs, making robots.txt legally significant, not just a convention. For non-AI scraping, robots.txt remains a strong good-faith signal that courts and regulators consider when evaluating whether collection was conducted responsibly.

What is the difference between legal scraping and ethical scraping?
Legal scraping means your collection method doesn't violate applicable computer access law or contract. Ethical scraping means your entire operation (including data type, collection method, infrastructure sourcing, and post-collection use) is defensible under the full range of applicable obligations, including GDPR, data minimization requirements, and your obligations to the individuals whose data you're touching. France's CNIL fined KASPR €240,000 in 2025 for collecting LinkedIn data that was arguably legal under CFAA analysis but non-compliant under GDPR, illustrating that legal and ethical are not the same standard.

What does GDPR say about scraping publicly available personal data?
GDPR applies to personal data based on whose data it is, not where the collector is based or whether the data was publicly accessible. Under Article 6, collecting personal data requires a documented lawful basis. For commercial scraping, "legitimate interest" is most commonly applied, but it requires a three-step proportionality test documented before collection begins. The European Data Protection Board's Opinion 28/2024 tightened this specifically for AI training, endorsing legitimate interest as possible but requiring strict proportionality analysis, and explicitly flagging that "public" does not mean "freely available for any purpose."

What is data provenance and why does it matter for proxy use?
Data provenance is the documented record of where data came from, how it was collected, through what infrastructure, and what has been done with it since. It matters for proxy use because the EU AI Act now requires general-purpose AI model providers to publish a training-data summary that documents sources and methods. This means data collected without provenance tracking creates retroactive compliance exposure. You may have collected it lawfully, but if you cannot document how, you cannot satisfy the disclosure requirement. Provenance also extends to infrastructure: data collected through a proxy network later linked to a botnet raises questions about the integrity of that data's collection chain.

What should an ethical proxy provider's acceptable use policy include?
An ethical AUP should include customer identity verification (KYC) before access is granted, explicit prohibition of credential stuffing, DDoS facilitation, PII harvesting without lawful basis, and scraping behind authentication; active traffic monitoring with enforcement mechanisms; and supply chain documentation showing how IPs in the pool are sourced and what consent mechanisms exist. A published AUP with no enforcement track record is not ethical infrastructure, it is a liability document. Enterprise buyers in 2026 are increasingly asking for evidence of enforcement, not just the policy text.

What are the ethical limits of using proxies for AI training data collection?
AI training data collection through proxies is subject to the full stack of data boundaries: you must have a lawful basis for any personal data fields, respect robots.txt and ai.txt opt-outs (legally required for EU AI Act compliance from August 2026), use ethically sourced proxy infrastructure whose provenance can be documented, and maintain records sufficient to populate a training-data summary if required. The EU AI Act's transparency requirements mean that data collected before provenance documentation was standard practice is harder to defend retroactively, making documentation at collection time essential for any organization building AI-ready datasets.

How do I know if my proxy infrastructure is ethically sourced?
Ask your provider directly: how does a device join your residential IP network, and what consent mechanism exists? Request a sourcing policy document. Reputable providers have them. Independently verify a sample of IPs via public ASN lookup (ARIN WHOIS, IPInfo): if the registered organization doesn't match the provider you're buying from, you're in a reseller chain, and the sourcing question applies to whoever is at the other end of it. For datacenter proxies, the sourcing question is simpler: verify the ASN, understand who shares your subnets, and ask about the pool's abuse history.