By Gustavo Becker, Executive Research Fellow, Berkeley Center for Law and Business
On November 5, 2015, the Fundão tailings dam collapsed above the village of Bento Rodrigues in Minas Gerais, Brazil. In minutes, a wave of mining waste destroyed the village. The toxic sludge traveled more than 650 kilometers down the Rio Doce to the Atlantic Ocean, contaminating drinking water and destroying the livelihoods of communities who depended on the river for fishing, farming, and daily life. The dam belonged to Samarco, a joint venture between the Brazilian company Vale and the mining group BHP. Beneath the environmental catastrophe lay a human rights question with no obvious answer: what was owed to the people whose homes, health, and livelihoods had been affected, and how would remedy be delivered? The following year, the companies, together with the oversight role of public institutions, created a private foundation to administer non-judicial remediation mechanisms based on mediation.
When the UN Human Rights Council endorsed the United Nations Guiding Principles on Business and Human Rights (UNGPs) in 2011, it called on companies to establish such non-judicial mechanisms. The UNGPs refer to them as grievance mechanisms, which are to be established in order to identify and address adverse human rights impacts. It framed them as a complement to judicial remedy, not a replacement for it. The legitimacy and promotion that the UNGPs provide for private remedial mechanisms is worth mentioning considering the Principles’ position as the authoritative global standard on business and human rights. They rest on three pillars: the State duty to protect human rights, the corporate responsibility to respect them, and access to remedy for those harmed. Grievance mechanisms sit in the third pillar.
Companies responded, and grievance mechanisms became corporate practice. For instance, Barrick Gold launched its Porgera Remedy Framework the same year the UNGPs were endorsed, private arbitration was used as a grievance mechanism in the aftermath of the Rana Plaza disaster in Bangladesh, and TotalEnergies today runs a grievance mechanism alongside the compensation and resettlement program for the East African Crude Oil Pipeline. Comparable mechanisms operate across extractives, apparel and agriculture.
These mechanisms may offer potential advantages for businesses and individuals: proximity to affected communities, speed, lower costs, transnational reach, and the capacity to deliver different types of reparations beyond financial compensation. My doctoral thesis identified a recurring problem among such mechanisms: the standards against which their effectiveness is analyzed.
The UNGP’s Effectiveness Criteria vis-à-vis an Effective Remedy under International Human Rights Law
Two ideas are easily conflated here, and the confusion has consequences. “Effectiveness,” under the UNGPs, refers to a set of design qualities that a grievance mechanism should exhibit. The “right to an effective remedy” is a right held by the individual and protected by human rights treaties. The first asks whether the process is well built. The second asks what the person was owed, and whether and how they received it. A mechanism can perform well against the first and still fall short of the second.
In business and human rights scholarship, almost all assessment of private remediation run through the effectiveness criteria of UNGP Principle 31. Principle 31 is the provision of the UNGPs that sets out what a non-judicial grievance mechanism should look like. It lists eight qualities: legitimacy, accessibility, predictability, equity, transparency, rights-compatibility, being a source of continuous learning, and, for company-managed mechanisms, being based on engagement and dialogue with the groups they serve.
These describe the qualities of a process. They do not necessarily encompass what remedy the claimant receives, whether the remedy provided matched the harm, or whether individuals could still go elsewhere if the process failed their interests. Principle 31(f) does require outcomes to accord with internationally recognized human rights, but without specifying which standards apply or what reparation they require.
The right to an effective remedy is more specific. It is set out in Article 2(3) of the International Covenant on Civil and Political Rights, which the United States ratified in 1992, and in every major human rights treaty. It generally requires both meaningful access to a process capable of addressing the violation and appropriate reparation for the harm suffered. On the first, the process must be reachable in practice, procedurally fair, and open to arguable claims, with access to an independent and impartial tribunal at its core. On the second, General Comment No. 31 of the UN Human Rights Committee and the case law of the Inter-American Court of Human Rights set out the forms that reparation may take: restitution, compensation, rehabilitation, satisfaction, and guarantees of non-repetition, proportionate to the gravity of the violation.
Such human rights treaties and their relevant case law constitute the sources of International Human Rights Law. These sources do not bind companies directly, and companies cannot be sued in a US court based on the existence of obligations under human rights treaties. They are sources of State-based obligations and may be considered benchmarks for corporate behavior, not a cause of action. Yet, they are the benchmarks a company should adopt when it attempts to meet international human rights standards. A company that endorses the UNGPs and then measures its mechanism only against Principle 31 is applying half of the framework it is supposed to.
The Business Interest in an Effectiveness Analysis
There is a strong business interest in the mentioned effectiveness analysis, particularly for corporate counsel. A private remedial mechanism that falls short of human rights standards may not close the claims. A mechanism providing inadequate remedies, or relying on legal waivers that may not hold, can fail to deliver the finality the company expects, exposing it to later litigation, additional compensation, and further legal costs.
Porgera illustrates the pattern. Eleven women represented by EarthRights International rejected the reparation offered through the private remedial mechanism established by Barrick Gold for victims of sexual assault conducted by the mine’s personnel around the mining site in Papua New Guinea. It is claimed that eligibility was confined to women who could prove sexual violence by mine personnel; compensation was standardized regardless of the severity of harm; and the final agreement signed between the mechanism and individuals required claimants to waive civil claims against Barrick in any jurisdiction. As their lawyers prepared to file suit, Barrick negotiated a settlement under which they reportedly received more than claimants who went through the private mechanism. No litigation was ever filed. Later that year, after complaints from women who had accepted packages and signed waivers through the private mechanism, Barrick announced an additional payment to those victims. The waivers did not prevent the company from paying again.
The Fundão dam collapse is another example. BHP was sued in relation to the incident in England. BHP argued that many claims in the English lawsuit were related to claimants who had already been compensated through the established private remediation mechanism through individual settlement agreements, many of which contained legal waivers and releases. In November 2025, in Município de Mariana v BHP Group (UK) Ltd, the English High Court found BHP liable under Brazilian law to more than 600,000 claimants, in a claim reported at around £36 billion. The Court did not accept that the releases barred the claims as a class. It held that they fall to be interpreted under the ordinary rules of the Brazilian Civil Code, that some of the sample agreements were adhesion contracts to be construed in the claimant’s favor where ambiguous, and that neither the judicial design of a scheme nor the assistance of a lawyer would preclude a finding that a particular agreement was invalid. Their effect on individual claimants falls to the second stage trial, listed for 2027. In May 2026 the Court of Appeal refused BHP permission to appeal on every liability and limitation ground.
Therefore, a release obtained in exchange for financial compensation in 2016 did not deliver the finality it appeared to promise, and by January 2026 the parties’ combined costs in the English proceedings had passed £300 million, before any damages had been assessed.
Four Questions to be Asked at the Design Stage
Eligibility: does the mechanism exclude categories of affected people, or impose evidentiary thresholds that filter out arguable claims before they are heard?
Control: are claim files administered by the company only or include external stakeholders? Are mediators independent? Are compensation levels set unilaterally?
Remedy: does the mechanism provide reparation that is proportionate to the harm and responsive to the claimant’s circumstances, rather than relying on standardized compensation alone?
Judicial access: does participation in private remedial mechanisms require a legal waiver to be signed by claimants?
This last question matters most, because access to an independent and impartial tribunal sits at the core of the right.
A Case for AI Automation of Procedural Checks
Whether a mechanism excludes categories of claimants, routes files to the company, or applies one standardized amount of compensation regardless of harm, procedural design might be assessed by AI-powered tools.
That is the direction of my current research at the Berkeley Center for Law and Business. The premise is that much of what the right to an effective remedy requires of effective remediation can be checked by AI tools. Was the complaint acknowledged, and when? Was the claimant told which standard would be applied to the claim? Did the outcome state reasons? Was reparation assessed against the criteria listed in General Comment No. 31 of the Human Rights Committee, or fixed at a standardized amount? Does the settlement paperwork contain a waiver? Each of these is a checkable data point in the mechanism’s records.
AI systems are suited to exactly this kind of work. Encoded with the procedural components of the right to an effective remedy, they could monitor a mechanism continuously rather than through periodic audits: flagging complaints filtered out at intake without reasons, tracking response times, confirming that every decision identifies the standard applied, detecting unequal treatment across claimant groups, and catching waiver language before it reaches a claimant.
The right to an effective remedy, by contrast, decomposes into requirements that are precise enough to be encoded. That is the practical payoff of adopting it as the benchmark: it turns remedy from an aspiration into a specification, and a specification can be automated.
Two limits frame the project. The first is legal. Article 22 of the GDPR restricts decisions based solely on automated processing where they produce legal or similarly significant effects for the person concerned. The outcome of a grievance is such a decision: it determines eligibility and the form and amount of reparation. The role for AI is therefore verification of the process, not adjudication of the claim. The judgment stays human.
The second is that automating any part of a grievance mechanism turns the mechanism itself into a regulated technology. In the EU, that means the AI Act. Its obligations arrive in phases, and the calendar has just moved. The Digital Omnibus on AI, in force since July 27, 2026, deferred the obligations for high-risk systems listed in Annex III from August 2, 2026 to December 2, 2027. As matters stand, the operative duty for most automated grievance tools is the transparency rule in Article 50, applicable since August 2, 2026: a claimant interacting with an AI system must be informed of it.
From December 2027, classification will determine how much more applies. A remedial mechanism whose outputs influence decisions affecting the employment relationship sits close to the high-risk category for employment in Annex III, which brings risk management, data governance, logging, human oversight, and, for some deployers, a fundamental rights impact assessment. A community-facing mechanism of the kind operated at Porgera or Mariana fits Annex III less obviously and may remain subject to little more than transparency. The EU regulates AI as a product, before it reaches the market, and coverage follows the product categories.
California starts from the other end. Under the CCPA regulations on automated decision-making technology, in effect since January 1, 2026, with full compliance due by January 1, 2027, a business using such technology for significant decisions, a category that includes employment decisions, owes the affected individual a pre-use notice, access to information about how the system works, and in most cases a right to opt out. Separate civil rights regulations impose anti-discrimination and record-keeping duties on automated-decision systems in employment. The handles sit with the individual rather than the regulator. For a US audience accustomed to treating Brussels as the stricter jurisdiction, the comparison is worth pausing on: for the person filing the complaint, California currently offers the more directly usable rights.
For corporate counsel, the convergence is the point. The AI Act’s high-risk obligations, from data governance to logging to human oversight, largely mirror what the right to an effective remedy already demands of the process: records, reasons, and a human responsible for the outcome. Subsequent posts in this series will take the components of the right to an effective remedy one by one, from access to timeliness to reasoned decisions to proportionate reparation to preserved judicial recourse, and ask for each what a machine can check, what it cannot, and what the governance frameworks on both sides of the Atlantic require of the answer.
The Bottom Line
Grievance mechanisms are useful for businesses and should be effective for individuals. The question is what standard they are built to. For a company that intends to implement the UNGPs, the right to an effective remedy is what gives that commitment content: it specifies what reparation must consist of and what access must be preserved.
For a company that treats the mechanism primarily as a cost-control device, the calculation is less reliable than it looks. Where claimants have no realistic route to a court, a narrow mechanism may hold. That is a bet on the continued absence of regulation, and states have begun to legislate. Paying twice should not be a risk to be taken by businesses.
About the Author:
Gustavo Becker is an Executive Research Fellow at the Berkeley Center for Law and Business and a Postdoctoral Researcher at the Luxembourg Centre for European Law. This post draws on his doctoral research at the University of Amsterdam and the University of Luxembourg, defended in January 2026.