ai risk today logo svg bl font
Menu
SUBSCRIBE
SUBSCRIBE
ai risk today logo svg bl font
Menu

EU AI Act

Exterior view of European Parliament building in Brussels with member state flags and EU emblem on the glass facade. Council approved changes to the EU AI Act.

EU Parliament Passes AI Act Changes, Pushing High-Risk Deadline to 2027

, , , , , , , , , ,

The 423-to-57 vote confirms deadline shifts and a ban on apps that generate non-consensual intimate images, pending Council adoption.

On Monday, the EU Parliament voted 423 to 57 to formally approve changes to the EU AI Act, confirming a series of compliance deadline extensions and a ban on apps that create non-consensual intimate images.

As AI Risk Today reported in May, EU lawmakers reached a provisional agreement on these changes under the European Commission’s Digital Omnibus simplification initiative. Monday’s vote completes Parliament’s approval. The law still requires formal Council adoption before taking effect.

 

Here’s what the vote confirmed – 

 

High-risk definition was narrowed. An AI feature that assists users or optimizes performance no longer automatically qualifies as high-risk. Only systems whose failure could pose health or safety risks are subject to those obligations. That removes a significant source of legal uncertainty for companies with embedded AI in their products.

“Nudifier” apps banned. Providers have until December 2, 2026 to either pull these systems from the EU market or add technical safeguards to prevent misuse. The ban also covers AI-generated child sexual abuse material.

Personal data can be used to detect bias. Companies gain a new legal basis to use personal data to detect and correct bias in their AI systems, provided they implement the required safeguards.

SME exemptions extended. Small and mid-cap companies get additional relief. The package extends SME exemptions to small mid-cap enterprises, which were previously excluded.

AI content watermarking confirmed. The watermarking requirement for AI-generated content, which will require a machine-readable label on AI outputs, takes effect the same day.

 

High-risk deadline pushed to December 2027

Standalone high-risk AI systems now have until December 2, 2027 to meet compliance requirements. Previously, the deadline was August 2026. AI systems embedded in products already covered by EU product safety laws have an additional year, until August 2, 2028.

 

The EU AI Act changes now await formal adoption by the Council. Most other AI Act provisions take effect August 2, 2026.

Satellite night view of Germany with Google's colorful logo overlaid, depicting the Munich court ruling on AI Overview defamation.

German Court Rules Google Directly Liable for AI Overview Defamation

, , , , , , , , , , ,

A Munich court ruled Google is liable for what its AI generates, not the sources it draws from.

 

A German court ruled that Google can be directly liable for false claims generated by its AI Overview. Munich’s Regional Court (Landgericht München I) found that AI-generated summaries are Google’s own content, not protected the way traditional search results are.

The case centered on Google’s AI Overviews that wrongly linked Verlagshaus24, a German publishing group, and a related subsidiary to scams, subscription traps, and questionable business practices. The claims were false: the AI connected them with entirely different, unrelated companies.

Verlagshaus24 notified Google in February 2026. Google did not remove the content. The two companies filed suit under German civil law and won on 8 of 10 claims.

 

Why Google lost

German courts have treated search engines as neutral middlemen responsible for removing harmful content only after being notified that it exists. The Munich court ruled that standard does not apply to AI Overviews.

The court found that Google’s AI Overviews added structure, drew inferences, and produced a specific claim presented as fact that was not traceable to any source it indexed. In the court’s view, that output was Google’s own statement, not a reflection of internet sources. Because Google controls the algorithm and the AI’s output, the court held Google directly responsible for the content.

 

What Google won

The court sided with Google on 2 of the 10 claims: that Verlagshaus24 “works with a debt collector” and “sells subscriptions improperly.” The court found that these claims were not sufficiently defamatory on their own, and that Verlagshaus24 did not demonstrate that they were false. Legal fees were assigned accordingly: Google pays 80%, the plaintiffs 20%.

 

This is the first German court ruling to address AI Overview defamation directly. The ruling aligns with the EU’s broader push to hold AI systems to the same standards as other content publishers. According to Reuters, Google said it will appeal. If the ruling stands, companies producing AI-generated summaries in Germany could face the same liability standards that apply to publishers, rather than the lower bar that has historically protected search platforms.

Futuristic city skyline with glowing data streams and network lines, from Check Point's 2026 Cloud Security Report.

77% Of Companies Changed Security Strategy For AI; Only 26% Can Enforce It

, , , , , , , , , , ,

Check Point’s 2026 Cloud Security Report finds more than half of companies already confirmed an AI-related security incident, and another quarter cannot rule one out.

Key Takeaways

  • 77% of companies changed their security strategy to manage internal AI use, but only 26% say their security systems can enforce those changes without a major rebuild
  • 54% confirmed at least one AI-related security incident. Another 24% suspect one but lack the monitoring to confirm it
  • Only 5% of security teams have full visibility into which AI tools employees use and where their data goes
  • Just 13% can block a malicious prompt before it reaches a model, and 5% can stop an AI response that leaks sensitive data or carries malicious content
  • 42% of organizations say employees bypass AI security controls when those controls slow them down

Check Point Software released its 2026 Cloud Security Report: Securing the AI Transformation, a study of how enterprise security is holding up as AI moves into production. Check Point surveyed 1,042 cybersecurity and IT professionals in early 2026, asking how AI has changed their security strategy, architecture, and day-to-day operations. The answers point in one direction: strategy is moving fast, and the infrastructure underneath it is not.

 

A 51-point gap between strategy and enforcement

AI adoption forced 77% of organizations to change how they manage security. As employees adopted tools like ChatGPT and Copilot, companies built AI into their own applications, security teams rewrote acceptable use policies, established AI governance boards, and redirected budgets toward AI-specific controls. But only 26% say their existing security systems can enforce those changes without a major rebuild.

The gap matters because AI is already running at scale. 70% of organizations run generative AI in production, 64% have AI agents in pilot or production, and 12% have given those agents privileged access to core systems. In plain terms: most companies decided what their AI rules should be, but few can apply those rules where AI actually operates.

 

More than half of companies report an AI incident, almost a quarter more suspect one

54% of organizations confirmed at least one AI-related security incident. Another 24% suspect an incident but lack the monitoring to confirm it. Combined, 78% either took a hit or cannot rule one out.

The most common incidents reported were:

  • Unauthorized or shadow AI use (41%)
  • AI-generated content used in an attack, such as phishing or deepfakes (37%)
  • Sensitive data leaked to or through AI services (32%)

 

Security teams see little and stop less

Only 5% of security teams have full visibility into which AI tools employees use, what data those tools touch, and where that data goes.

Enforcement is not much better. 13% can block a malicious prompt before it reaches a model. 16% can stop sensitive data from reaching AI services in real time. Just 5% can reliably stop an unsafe AI response before it reaches users or other systems. Most tools can raise an alert after the fact but cannot stop the activity in real time.

Weak enforcement gives employees a way around the rules. 42% of organizations say employees bypass AI security controls when those controls slow them down, often by pasting data into a personal AI account or using a browser-based tool the company’s security software does not cover.

 

Companies are shifting security budgets to close the gap

52% of organizations are increasing their AI security budgets, and 37% are replacing their collection of security tools with a single platform, nearly double the 20% still buying a separate tool for each problem. Regulation may force the pace: only 7% say they are fully prepared for AI-specific rules such as the EU AI Act and new state laws in Illinois and Colorado.

OpenAI logo on an orange and blue gradient background, accompanying coverage of the OpenAI Frontier Governance Framework release.

OpenAI Releases Safety Governance Document Aligning With Emerging Regulations

, , , , , , , , , , ,

The 20-page Frontier Governance Framework details risk thresholds that determine whether a model can be deployed, tying each decision to California and EU law.

 

OpenAI published its Frontier Governance Framework, a 20-page document that outlines how the company identifies, evaluates, and responds to the most dangerous risks its models could pose. It covers four categories: cyber offense, CBRN (chemical, biological, radiological, and nuclear weapons), harmful manipulation, and loss of control. Loss of control covers models that could deceive humans, evade shutdowns, or take unauthorized autonomous action.

The release outlines how OpenAI’s security guidelines align with current and future regulations, including California’s Transparency in Frontier AI Act and the EU AI Act. Both examples require OpenAI to document its processes for managing catastrophic risks from its most capable models.

 

What triggers a risk review

OpenAI defines a systemic risk as any threat that could materially contribute to more than 50 fatalities or $1 billion in property damage from a single incident. For each of the four risk categories, the company uses a three-tier system to measure how much a model raises the danger level.

Tier 1 covers model capabilities comparable to other publicly available methods of gathering information. For example, if someone could get the same help by searching the web, reading a manual, or asking a domain expert, the model is operating at Tier 1. Tier 2 is when a model provides meaningful uplift to bad actors who couldn’t otherwise achieve the same result. For example, enabling someone with basic technical knowledge to create a known biological or chemical threat. Tier 3 covers the most severe: in the CBRN category, that means a model capable of enabling an expert to develop a novel pathogen, or one that could autonomously execute the full design and production of a biological weapon without human involvement.

If a model’s residual risk exceeds acceptable levels after mitigations are applied, OpenAI says it will not deploy it.

 

What’s still under development

Regarding loss of control, the risk tiers are still being developed, except for risks related to AI self-improvement, for which the thresholds are more clearly defined.

For harmful manipulation, OpenAI has not yet developed a way to evaluate risk before deployment. Risks in that category, including influence operations and election interference, are currently addressed through post-deployment monitoring rather than pre-release testing.

 

Reporting reviews and guardrail changes

The company reviews safety reports for covered models at least every six months. The framework itself gets a full assessment at least annually. Material changes require board-level approval at the OpenAI Foundation and must be published within 30 days.

 

OpenAI will update the framework as regulatory requirements and model capabilities evolve.

Two professionals at laptops in a glass high-rise office at night, city skyline behind them, reviewing AI safety benchmarks results.

Every Frontier AI Model Tested Fails Under Real Attack Conditions, Cisco Finds

, , , , , , , , , , ,

Cisco tested 15 leading AI models using multi-turn attacks, the way real attackers operate, and found attack success rates as high as 88%.

When AI companies test their models for safety, the standard approach is one challenge at a time: an attacker sends a single harmful message, the model responds, and the test ends. This is called a single-turn prompt or single-turn attack. This is how AI safety benchmarks evaluate and compare AI models today. Real cyberattackers do not work that way. They probe, reframe when blocked, and keep pushing back and forth, turn after turn, until they find a way through. A test that mirrors that kind of sustained, adaptive pressure is called a multi-turn attack. Cisco ran that test across 15 of the most widely deployed AI models from every major provider; not one held up.

Key Takeaways

  • Multi-turn attack success rates reached 88.30% for Grok 4.1 Fast when its reasoning feature was turned off, the highest among the 15 models.
  • Every model tested showed meaningful vulnerability to multi-turn attacks. The lowest in the group was Amazon Nova 2 Lite at 7.89%.
  • GPT-5.4 went from a 2.74% single-turn attack success rate to 24.68% under multi-turn testing, a 9x increase.
  • Gemini 3 Pro rose from 18.10% to 73.35% under multi-turn conditions, a jump of 55 percentage points (pp).
  • Turning on Grok 4.1 Fast’s reasoning feature cut its multi-turn attack success rate from 88.30% to 43.47%.
  • 8 of 15 models showed a gap greater than 15 pp between single-turn and multi-turn success rates.

Cisco’s AI Threat Intelligence and Security Research Team published the Proprietary Problems report on May 27. The report ran 15 leading AI models from OpenAI, Anthropic, Google, Amazon, and xAI through 30,090 single-turn prompts and 6,986 multi-turn attack sequences. The two types of tests produced different rankings, risk levels, and implications for how companies choose which model to deploy.

 

The benchmark problem

Safety testing organizations typically employ single-turn testing because it targets specific weak points, allowing companies to focus on and fix them individually. Multi-turn testing accounts for how attackers behave and emulates that behavior. Here, the idea is to see what methods work, not to test individual vulnerabilities. Real attackers reframe when blocked, break requests into smaller pieces spread across multiple exchanges, and escalate until they find a way through.

Cisco’s test ran both approaches side by side. Multi-turn attack success rates across the 15 models ranged from 7.89% to 88.30%, a range 18 pp wider than the single-turn spread of 2.19% to 64.91%. Eight of the 15 models showed a gap of more than 15 pp between the two test types.

 

No model is clean

Every model in the group showed vulnerability to multi-turn attacks. The lowest multi-turn attack success rate in the study was Amazon Nova 2 Lite at 7.89%. The highest was Grok 4.1 Fast at 88.30%, tested with its reasoning feature turned off. Most of the group fell between 11% and 31%.

 

Single-turn scores hide risk

The single-turn numbers give a misleading picture for several models. GPT-5.4 has a 2.74% single-turn attack success rate, second-lowest in the group. Under multi-turn testing, it reaches 24.68%, a 9x increase. GPT-5.2 moves from 4.74% to 23.50%. Gemini 3 Pro starts at 18.10% and climbs to 73.35%, a 55 pp jump. Those shifts don’t appear on standard safety tests.

The Amazon models moved in the opposite direction. Nova Lite’s attack success rate dropped by almost 35 pp between single-turn and multi-turn testing; Nova Micro’s dropped by 34 pp. 

Grok 4.1 Fast reached an 88.30% multi-turn attack success rate when its reasoning feature was turned off. Reasoning mode is a setting that causes an AI model to work through a request step by step before responding, rather than answering immediately. With that setting turned on, the same model dropped to 43.47%, a reduction of nearly 45 pp from one configuration change. Cisco argues that AI providers should tell customers which settings meaningfully affect security, not just performance.

AI Risk Today AI Safety Benchmarks

(Performance by model –  Source: Cisco Proprietary Problems: How Frontier Closed Models Collapse Under Iterative Pressure)

 

What to do about it

Cisco recommends three steps for frontier AI developers to consider as part of the development and deployment process. 

  1. Publish attack success rates, broken down by attack type, for every release, not just a single overall number. 
  2. Hold a model back from deployment if it gets worse on the three highest-risk attack approaches (Imposter AI, Soft Paraphrase, and System Prompts) or the three highest-risk content categories (Hate Speech, Profanity, and Specialized Advice). 
  3. Three, flag any model showing a gap greater than 15 pp between single-turn and multi-turn attack success rates for manual review before deployment, a threshold that would have flagged eight of the 15 models in this study.

 

Governance and regulation considerations

NIST’s AI Risk Management Framework and its forthcoming Cyber AI Profile both call for attack-based testing of AI models. The EU AI Act’s Article 15 requires robustness testing for high-risk AI systems starting December 2027. Neither currently specifies multi-turn testing or the attack-type breakdown Cisco advocates.

Digital European Union flag representing EU AI Act high-risk AI system guidance and regulation

EU Draft Guidance Defines What Counts As High-Risk AI

, , ,

The European Commission’s draft guidance gives businesses a clearer view of which AI systems will face stricter EU compliance rules and which likely will not.

 

Key Takeaways

  • Companies are not classified as high-risk. Specific AI systems and use cases are.
  • AI systems are most likely to become high-risk if they affect safety, critical decisions, or people’s rights.
  • Using AI in hiring, healthcare, transportation, education, law enforcement, or critical infrastructure can trigger high-risk classification.
  • Most general business AI tools, chatbots, productivity systems, and consumer convenience products likely will not qualify as high-risk.

 

The European Commission released draft guidance explaining how regulators and businesses should determine whether an AI system qualifies as high-risk under the EU AI Act. The guidance is intended to help companies understand when stricter compliance obligations apply.

The guidance points out that the AI Act does not label entire companies as high-risk. Instead, it classifies specific AI systems, products, and use cases within a business. A company could operate dozens of AI systems internally while only one or two fall into the high-risk category.

If it affects human safety, it’s high-risk

Under the draft guidance, AI systems are likely to be high-risk when they can significantly affect people’s health, safety, legal rights, or access to important services.

AI systems connected to products currently regulated under EU safety laws would be considered high-risk – for example, medical devices, industrial machinery, automobiles, and aviation systems.

If the AI system performs a safety-related role within those products, it may be classified as high-risk.

Some examples include:

  • Detect humans near industrial robots
  • Monitor gas leaks
  • Prevent train collisions
  • Trigger emergency shutdowns

The Commission said these systems qualify because a malfunction could physically harm people or property.

 

And if it affects important decisions that affect people’s lives, it’s high-risk

The EU’s core concern is whether the AI system can significantly affect a person’s economic opportunities, safety, or legal rights.

In practice, this means AI systems are more likely to be classified as high-risk if they influence decisions related to employment, financial decisions, healthcare, education, law enforcement, or access to services.

However, most low-impact business AI tools, such as chatbots, meeting summaries, drafting tools, and marketing automation systems, likely would not qualify as high-risk unless they are directly tied to high-impact decision-making.

 

Your marketing materials and product positioning could matter

The Commission also made clear that classification depends heavily on how a company describes and promotes its AI systems. According to the draft guidance, regulators may review product and marketing information, including forward-facing sales and marketing materials, technical documentation, product instructions, and public statements to determine the status.

The Commission warned that actual functionality and real-world positioning matter more than broad disclaimers attempting to distance the company from high-risk use cases.

 

Office productivity and consumer AI tools are less likely to be classified high-risk

Most consumer and work productivity systems likely fall outside the category unless they create safety or material risks. Home appliances, virtual assistants, and smart thermostats are just a few examples.

The Commission said these systems can cause inconvenience, financial loss, or user frustration when they fail, but are limited in what physical and material harms they can cause. 

 

Products that require third-party certification may attach additional AI obligations

The guidance also ties high-risk classification to existing non-AI EU product safety statutes.

If a product already requires third-party safety certification under EU law, the embedded AI system may automatically incur additional obligations under the AI Act.

For those already adhering to product safety laws, e.g., manufacturers, healthcare companies, transportation firms, and infrastructure providers, this could significantly increase compliance requirements, including documentation, monitoring, and regulatory reporting. 

 

The draft guidance is still in draft

The Commission stressed that the document is still in draft form and is open to stakeholder consultation before final adoption. The Commission also said additional sector-specific guidance and examples are expected as implementation continues.

The guidance comes on the heels of the Commission’s recent revisions to the AI Act’s digital Omnibus package, which delayed enforcement of the AI Act.

The guidance is not legally binding. Final interpretation of the AI Act ultimately rests with the Court of Justice of the European Union.

European Union flag with digital overlay representing EU AI Act changes, including delayed high-risk AI rules and new restrictions on AI-generated explicit content

EU Lawmakers Reach Deal To Ease AI Act Requirements

, , ,

The provisional agreement postpones the enforcement of regulations on AI systems deemed high risk while adding new restrictions on AI-generated sexually explicit content.

 

EU country representatives and parliament legislators agreed to delay enforcement of the EU AI Act under the bloc’s “Digital Omnibus” simplification initiative, according to an official statement from the European Parliament. The agreement also eases restrictions on manufacturing companies while adding a ban on AI systems that generate unauthorized sexual images.

Here’s what changed – 

1. High-Risk AI System Rules
Before: Requirements for “high-risk” AI systems were slated to take effect August 2, 2026. “High-risk” systems are defined as those that could pose health and safety risks or endanger basic rights. AI developers would be required to implement risk management systems, maintain technical documentation, ensure human oversight, and meet testing and monitoring obligations.

After: The agreement delays application of these requirements until December 2, 2027. Notably, manufacturing companies will, for the most part, be exempt from these requirements as existing EU product safety legislation already imposes requirements.

 

2. Watermarking requirements for AI-generated content become enforceable starting December 2, 2026, moving up from February 2, 2027.

 

3. “Nudifier” and child porn AI apps banned. AI systems that generate child pornography or undress identifiable people without consent will be illegal and enforced starting December 2nd, 2026.

 

The changes are part of the European Commission’s broader effort to simplify EU digital regulations after businesses and some member states argued overlapping compliance requirements created legal uncertainty and excess administrative red tape. Negotiations had previously stalled over whether AI systems already regulated under sector-specific laws should also remain subject to the AI Act.

The provisional agreement requires formal approval from the European Parliament and EU member states before taking effect.

European Union flags displayed outside an EU institutional building in Brussels

EU Council Backs Ban On AI Nudification Tools In EU AI Act Update

, ,

The proposal would prohibit AI systems that create non-consensual nude or sexually explicit images as part of the “Omnibus” amendments to the EU AI Act.

 

The Council of the European Union agreed to add a ban on certain AI systems capable of generating non-consensual nude or sexually explicit images, often referred to as “nudification” tools, as part of ongoing amendments to the EU AI Act.

According to the Council’s official release and legislative text, the proposal would prohibit AI practices that generate “non-consensual sexual and intimate content.”

The restriction would apply to developers, providers, and users of AI systems operating within the European Union, including those that place AI tools on the market or make them available to users. The measure targets systems that manipulate or generate realistic images depicting individuals in sexual or intimate contexts without their consent.

The Council stated that the amendment is intended to address gaps in the current law, which does not explicitly ban such tools.

In its official statement, the Council said the updated framework is designed to “streamline the application of the rules” while maintaining protections tied to safety and fundamental rights.

The Council’s position now serves as its negotiating mandate in discussions with the European Parliament, which is considering a similar, less specific prohibition under which “nudification” tools would be banned under the “prohibited practices” (Article 5) section on non-consensual intimate AI-generated content.

The proposal has not yet been adopted into law. Any final changes to the AI Act will require agreement between the Council and Parliament before entering into force.

Axel Voss, Member of the European Parliament, at the European Parliament in Strasbourg

EU Lawmakers Back New Generative AI Copyright Protection Proposals

, , , ,

The measure would require transparency and fair pay for the use of copyrighted works in generative artificial intelligence training.

Members of the European Parliament’s Legal Affairs Committee approved a set of proposals, “Copyright and generative artificial intelligence – opportunities and challenges,” aimed at protecting copyrighted works used by generative AI. The proposals state that creators should have transparency, control, and remuneration for the use of their material in AI training data.

The proposals would require that European Union copyright law apply to all generative AI systems available on the EU market, regardless of where the training took place, and call for full transparency about the copyrighted content used in training, including lists of works and records of data crawling by AI providers and deployers.

The committee text also seeks fair remuneration for rights holders for the use of copyrighted material by generative AI and urges the European Commission to explore whether such payments should cover past uses. It rejects the concept of a single global licence with a flat-rate payment for AI training.

In its report, the committee says news media and individual creators should be able to refuse use of their content for AI training and calls on the Commission and member states to act to protect media pluralism and ensure remuneration for use of press content.

The committee text further states that works fully generated by AI should not qualify for copyright protection and calls for measures to address the dissemination of manipulated AI-generated content, as well as obligations on digital service providers to counter illegal use.

After the committee vote, Voss said, “Generative AI must not operate outside the rule of law. If copyrighted works are used to train AI systems, creators are entitled to transparency, legal certainty, and fair compensation.”

The report will be put to a vote in the full European Parliament plenary, expected in March.

EU Launches Confidential Whistleblower Tool to Enforce AI Act

EU Launches Confidential Whistleblower Tool To Enforce AI Act

, , ,

The European Commission has launched a new confidential reporting channel for suspected violations of the EU Artificial Intelligence Act, introducing what it calls the AI Act Whistleblower Tool. The mechanism is designed to help regulators identify potential breaches early, particularly when employees, contractors, or other insiders are the first to detect non-compliant or harmful AI practices. The Commission formally announced the tool on November 24.

According to the Commission, the reporting system allows individuals to submit information about suspected violations involving general-purpose AI models or other AI systems covered by the Act. Reports may concern risks to fundamental rights, health, safety, or failures to meet legal obligations under the AI Act. Submissions can be made anonymously, in any official EU language, and are open to EU employees, citizens, and other relevant parties.

The tool is operated by the European Artificial Intelligence Office, the body responsible for overseeing compliance with the AI Act at the EU level. The Commission said the system uses encrypted communication channels and allows whistleblowers to upload supporting materials such as documents, technical information, or other evidence. A secure two-way inbox enables the AI Office to ask follow-up questions without revealing the reporter’s identity. The Commission stated that no identifying metadata is collected.

In guidance accompanying the launch, the Commission advises individuals not to use employer-owned devices or networks when submitting reports and to avoid including information that could indirectly identify them. Examples of relevant submissions include internal documentation, audit materials, technical specifications, code samples, and descriptions of observed AI system behavior.

While the reporting tool supports anonymity, formal legal protections against retaliation under the EU Whistleblower Directive will apply to AI Act–related disclosures starting in August 2026. Until then, confidentiality is the primary safeguard. The Commission noted that some AI-related disclosures may already qualify for protection under existing whistleblower rules, including those involving product safety, consumer protection, cybersecurity, and data protection.

The launch comes as the EU continues to build its enforcement framework for the AI Act, which entered into force in August 2024. The law imposes obligations on high-risk AI systems, sets requirements for general-purpose AI models, and prohibits specific AI uses deemed incompatible with fundamental rights. The Commission said confidential reporting will support the AI Office’s oversight work and help authorities prioritize enforcement actions, with relevant cases shared with national regulators where appropriate.

Load More