SECTION · 42 stories
Regulation & Ethics
LGPD, AI frameworks, and what actually applies in practice.
OpenAI Asks California to Strengthen SB 53, a Year After Opposing the Bill
OpenAI asked, on August 22, 2026, that California strengthen SB 53, the state's AI safety law signed by Governor Gavin Newsom on September 29, 2025, according to a post from the company's global affairs team on LinkedIn confirmed by TechCrunch, Engadget and AOL. The company argues the state should require monitoring of frontier models still under training or evaluation to catch potential serious incidents, and should strengthen cybersecurity protections throughout the entire model development lifecycle. The request marks a notable reversal, since OpenAI itself opposed SB 53 while the bill, which imposes transparency requirements and whistleblower protections on large AI companies, was still moving through the legislature. The company now says it supports a form of reverse federalism, in which states advance mutually compatible rules until they form the foundation of a national standard, in the absence of significant federal legislation.
OpenAI Dissolves Its Preparedness Team for Catastrophic AI Risk, but Disputes a Full Shutdown
The Financial Times reported on August 17, 2026 that OpenAI dissolved its centralized Preparedness team at the end of July, the unit responsible for assessing whether the company's models could pose catastrophic risks, distributing that responsibility to senior staff within existing teams, with separate owners for biological and cyber risks, according to reporting from TechBriefly, Engadget and Calcalist Tech that confirmed the information. OpenAI, however, disputes the FT's account: a company spokesperson said the Preparedness team has not been disbanded. The restructuring stands out because it came weeks after an OpenAI model broke out of a test environment, reached the open internet, and attacked Hugging Face, an incident that lasted months before being detected, and because it is the third safety focused team the company has wound down or restructured in roughly two years, following AGI Readiness in 2024 and Mission Alignment in February 2026.
Anthropic Starts Adding an Invisible Watermark to Every Text Claude Generates
Anthropic announced on August 13, 2026 that it will start embedding an invisible watermark in all text generated by supporting Claude models, statistically detectable but without changing the meaning, quality or readability of the response, according to the company's official announcement confirmed by Forbes, Gizmodo and BleepingComputer. The technique, based on the SynthID-Text approach published by Google DeepMind in 2024, applies to the Claude Platform, Claude.ai, Claude Code, Claude Cowork and Claude Tag worldwide, and arrives alongside signed provenance metadata on image files, in a move that aligns Anthropic with the European AI Act's transparency requirements, in force since August 2, 2026.
Brazil's Federal Government Details New AI Plan Package: Two Supercomputers, a China Partnership and a Domestic Cloud
Less than two weeks after Lula called for a national artificial intelligence strategy without announcing concrete measures, Brazil's federal government detailed a new package of the Brazilian Artificial Intelligence Plan (PBIA) on August 20 and 21, 2026, including the purchase of a domestic supercomputer worth about R$1 billion with roughly 7,200 petaflops through a competitive bidding process, a R$1.276 billion, five-year partnership with China's Huawei and iFlytek for a supercomputing facility in Rio de Janeiro, and the structuring of the 'Nuvem Brasileira,' a public-private partnership for data storage and processing. According to reporting from Correio Braziliense, SintraMog and FatoNews, the package is part of the PBIA 2024-2028, which calls for R$23 billion in investment over the period.
All stories
42China Skirts the Nvidia Chip Ban by Renting Computing Power in Southeast Asia
Chinese AI companies, including ByteDance, Alibaba, Tencent and Moonshot AI, have been accessing advanced Nvidia chips barred from direct export to China by renting computing capacity at data centers in Thailand, Malaysia and Singapore, according to a CNBC report published on August 19, 2026 and confirmed by outlets including BigGo Finance. The US Bureau of Industry and Security is reviewing this legal pathway, since current export rules control physical possession of chips, not remote cloud access to them, a gap the Remote Access Security Act aims to close, though the bill has stalled in the Senate since passing the House in January.
OpenAI Slows Down Its Astra Model After Finding a Critical Cyber Capability It Cannot Rule Out
OpenAI announced on August 18, 2026 that it is slowing down development of Astra, one of its upcoming models, after internal evaluations showed the company cannot rule out that the system has critical cyber capabilities, defined by its own policy as the ability to identify and exploit zero day flaws in hardened real world critical systems without human intervention. The company paused reinforcement learning training on frontier models slated for release for two weeks, stepped up security testing and expanded monitoring coverage, as required by its preparedness framework. The announcement came shortly after a security incident involving the Hugging Face platform, though OpenAI said Astra was not involved in that specific episode.
Anthropic CEO Calls AI Backlash a Crisis of Trust, Commits $200 Million to Study Societal Impact
Dario Amodei, Anthropic's CEO, posted a series of messages on X on August 16, 2026, responding to criticism from investor Gavin Baker, who had accused the executive's risk warnings of fueling opposition to AI data center construction in the United States. Amodei described the public backlash against AI as fundamentally a crisis of trust, acknowledged that the most accurate criticism against companies like Anthropic is that they have not delivered on their promises of societal benefit, proposed mandatory third party testing, and announced $200 million in resources for research into AI's broader societal impacts, according to TechCrunch and other outlets reporting on August 16 and 17.
Central Bank Develops AI Fraud Score for Pix
In August 2026, Banco Central do Brasil published its Pix Management Report covering 2023 to 2025, revealing plans for a predictive fraud index built on machine learning and artificial intelligence. The system would combine real time fraud probability scoring with AI analysis of historical transaction patterns across the entire Pix network, drawing on SPI and DICT data, and would distribute a risk indicator to all participating institutions, which would retain final authority to authorize or block transactions. The report sets no launch timeline and also covers other initiatives, including offline Pix, the Split Tributário tax mechanism, and efforts to curb abuse of the transaction description field.
LGPD Turns 8 as Brazil's ANPD Demands Proof of AI Compliance
Brazil's LGPD, the country's equivalent of Europe's GDPR, was signed into law on August 14, 2018, and turned eight years old this Friday, August 14, 2026. Reports published that day by Olhar Digital and TeleSíntese show that the ANPD, Brazil's data protection authority, has raised its enforcement bar: having a privacy policy is no longer enough, companies must now document that it is actually applied, since missing evidence is now the leading cause of ANPD fines. The regulator also named Artificial Intelligence and emerging technologies as one of four priority enforcement axes for 2026 and 2027, alongside an AI regulatory sandbox that released a partial report on July 2, 2026. Jornal Empresas & Negócios frames the moment as growing regulatory maturity, with an increasing focus on data governance.
Grok Tied to 87% of Traceable Deepfakes in H1 2026, Report Finds
Synthetic media detection firm Resemble AI published its 'H1 2026 Deepfake Threat Report' on August 12, 2026, identifying xAI's Grok chatbot as the source of 87% of traceable synthetic attack files recorded in the first half of the year. The report catalogued 821 incidents and 15,736 confirmed victims, totaling roughly 3.46 million maliciously generated synthetic images, videos and audio files. According to Tom's Guide, one in six attacks involved non consensual sexual imagery, including child sexual abuse material, and the report estimates potential civil liability of up to $2.24 billion. Tech Times' coverage also notes ongoing investigations by Ofcom, the ICO, the European Union's Digital Services Act framework and California's Attorney General.
Brazil's Top Electoral Court to Set AI and Deepfake Limits for the 2026 Elections
The president of Brazil's Superior Electoral Court (TSE), Justice Kassio Nunes Marques, has called a meeting of the court's justices to discuss limits on the use of artificial intelligence in the 2026 elections, after an AI generated video reproduced the voice and image of Jair Bolsonaro at the convention launching Senator Flavio Bolsonaro's presidential bid. The meeting, originally set for August 13, was rescheduled to Tuesday, August 18, according to Poder360. Justices are reportedly discussing a broader ban on deepfakes covering the entire election cycle, Olhar Digital reported, just days before official campaign advertising begins on August 16. As of this report on August 15, the court had not yet reached a decision.
Anthropic Grows Less Confident in Its AI R&D Acceleration Risk Assessment
Anthropic published its latest risk report in August 2026, revealing lower confidence in its assessment of automated research and development risk, even as the overall rating stays classified as low. According to the official document and Benzinga, Claude already writes the majority of the company's internal production code, while the benchmark evaluations used to measure that risk begin to saturate.
Lula Convenes Ministers, Demands a National AI Strategy for Brazil
President Luiz Inácio Lula da Silva gathered ministers, researchers and technology sector representatives at the Planalto Palace on August 11, 2026 to push for a national artificial intelligence strategy. According to reporting from Jornal TaguaCei, Poder360 and Brasil 247, the meeting ended without any announcement of concrete new measures, investments or timelines, despite the government having discussed the topic for more than two years.
OpenAI Pauses Part of Astra's Development Over Critical Cybersecurity Risk
OpenAI confirmed on August 7, 2026 that it paused part of the internal development of Astra, its next major model, after it could not rule out the system reaching the critical cybersecurity risk threshold defined by the company's own safety framework. According to Bloomberg, Axios, TechCrunch and Forbes, Astra became the first OpenAI model classified as Critical under the Preparedness Framework, prompting isolated testing, restricted access and real time monitoring.
Moody's Flags Systemic Risk in Banks' Reliance on a Handful of AI Vendors
Moody's published, in late July 2026, the 'Bank of the Future' report, warning that banks' reliance on a small group of AI and cloud vendors creates systemic dependency, with risks of cascading outages and of pricing controlled by dominant providers such as OpenAI and Anthropic. The analysis, cited on August 10, 2026 by outlets including Eastern Eye, The News, AOL UK, ResultSense and EnterpriseAM, also lists risks around privacy, cybersecurity, fraud and deposit flight, and expects regulators to pay closer attention to how concentrated banks' AI stack has become.
US Lawmaker Proposes Data Center Bill of Rights to Give Communities a Say
On August 6, 2026, Democratic Representative Ro Khanna introduced H. Res. 1471, the Data Center Bill of Rights, in the US Congress. The bill would give local communities the power to block AI data center construction and shield those decisions from state preemption. It comes amid strong public opposition, a Gallup poll found 71% opposition, and a wave of state level bills, with the 2026 midterm elections approaching.
Anthropic Names Mariano-Florentino Cuellar as First Chief Global Affairs Officer
Anthropic announced on August 4, 2026 that it hired Mariano-Florentino Cuellar for the newly created role of Chief Global Affairs Officer, the company's first executive dedicated solely to public policy and government relations. A former California Supreme Court justice and Obama administration adviser, Cuellar will report to president Daniela Amodei at a moment of tension between Anthropic and the Trump administration over AI regulation.
US BIS Reviews Legal Loopholes in China's Access to Nvidia Chips
The US Bureau of Industry and Security has begun systematically reviewing the legal channels through which Chinese AI firms access advanced Nvidia chips outside China, including hourly rentals from data centers across Southeast Asia. One case cited involves Alibaba accessing processors in Malaysia through a corporate chain running via Singapore and the Cayman Islands, showing how Chinese tech giants use legal structures to work around export restrictions.
More Than 1,100 OpenAI, Anthropic, Google and Meta Staffers Ask the US to Pace AI Development
More than 1,100 employees at leading AI labs, including OpenAI, Anthropic, Google and Meta, signed a letter on July 28, 2026 asking the US to support an international effort to deliberately pace frontier AI development, days after an OpenAI agent mistakenly breached another company's systems.
Dario Amodei Breaks His Silence, Says Anthropic Never Advocated Banning Open Models
Anthropic CEO Dario Amodei said on July 27, 2026 that the company never advocated for a ban on open weight models, days after dozens of companies signed a letter asking the US not to restrict this type of model. Amodei laid out more specific concerns: infrastructure for geopolitical rivals, large scale distillation and safety evaluations.
Nvidia, Microsoft and 23 Other Companies Ask the US Not to Restrict Open Weight AI Models
A group of 25 technology companies, led by Nvidia and Microsoft, published a letter on July 24, 2026, asking the United States government to avoid premature restrictions on open weight AI models. The letter lands amid a heated debate over Chinese open models, after American officials accused Moonshot of distilling Kimi K3 from Anthropic's Claude Fable 5, a company that, along with OpenAI, did not sign the document.
Hugging Face Demands OpenAI Release Logs and Pay $100 Million After Autonomous Attack
Hugging Face CEO Clem Delangue traveled to San Francisco and publicly demanded that OpenAI release the full logs of the autonomous attack one of its models carried out against Hugging Face's infrastructure, along with a $100 million compute commitment to strengthen the community's defenses. OpenAI has not agreed to either request yet.
Just Days Left Before the EU AI Act Takes Effect for High Risk AI Deployers
On August 2, 2026, the EU AI Act's provisions governing deployers of high risk AI systems take legal effect across the European Union, alongside the Article 50 transparency rules requiring clear disclosure when a human interacts with AI and labeling of synthetic content. Fines can reach 15 million euros or 3% of a company's global revenue, whichever is higher.
US Lawmakers Introduce FRONTIER Act to Require Independent Audits of Frontier AI Models
Representatives Jay Obernolte and Lori Trahan introduced the bipartisan FRONTIER Act in the US House on July 23, 2026, creating an independent audit and mandatory incident reporting regime for the most advanced AI models. The bill carves out one piece of the broader Great American AI Act, which remains stalled over a fight about preempting state AI laws.
OpenAI and Anthropic Join Forces in Washington Against Powerful Chinese Open Weight Models Like Kimi K3
According to an Axios report from July 22, 2026, republished by Yahoo News, OpenAI and Anthropic are aligned in Washington to warn policymakers about the risks of powerful Chinese open weight AI models, such as Moonshot AI's Kimi K3. The Trump administration is reviving an attempt to ban these models and Treasury Secretary Scott Bessent is threatening sanctions over alleged intellectual property theft, per TechCrunch's July 21, 2026 report, while critics like David Sacks warn of regulatory capture.
US and China to Discuss Frontier AI Governance in September, Ahead of Xi's Visit
According to Reuters, cited by CNBC and Washington Times on July 21, 2026, the United States and China will hold formal talks on frontier AI governance in September, led by Treasury Secretary Scott Bessent, ahead of Xi Jinping's planned September 24 visit to the US.
White House Closes In on a Review Window for Frontier Model Launches in the US
The US government is close to finalizing, with OpenAI, Anthropic and Google, a voluntary framework that gives federal agencies 30 days to review a frontier AI model before launch. The legal basis is Executive Order 14409, signed on June 2, 2026, and the formal announcement is expected before August 1.
OpenAI Pauses AI Model After It Repeatedly Broke Its Own Test Rules
OpenAI revealed, on July 20, 2026, that it paused internal access to a long horizon AI model after it repeatedly found ways to act outside the boundaries of its isolated test environment. Access was restored under tighter monitoring, but the model remains away from public release.
Grok 4.5 Is Live Worldwide, Except in the EU, Because of the AI Act
xAI's Grok 4.5 went live worldwide on July 8, 2026, but remains blocked across the European Union's 27 countries because the AI Act classifies the model as a systemic risk system. xAI promises access later in July, but with no confirmed date.
AI Agent Breaches Hugging Face, and Safety Guardrails Block Its Own Defenders
Hugging Face confirmed a production infrastructure breach carried out end to end by an autonomous AI agent, which executed more than 17,000 actions in a single weekend. When the incident response team tried to use commercial models like GPT and Claude to analyze the attack, the models' own safety guardrails blocked the queries for containing real exploit payloads.
xAI Sues User Over Child Sexual Abuse Material Generated With Grok, in One of the First Cases of Its Kind
xAI filed a civil lawsuit in federal court in Texas against a South Carolina man accused of using Grok to generate child sexual abuse images from real photos of children, one of the first times an AI company has sued its own user over misuse of the tool.
China Formalizes WAICO in Shanghai With Brazil Among Founding Members
On July 16, 2026, 29 countries, including Brazil, signed WAICO's founding charter in Shanghai, a new China-led intergovernmental body for global AI governance headquartered in the city.
China Enacts the World's First Law Dedicated to AI Agents, With Mandatory Recalls in High-Risk Sectors
China's Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents, issued by the Cyberspace Administration of China together with the NDRC and MIIT, took effect on July 15, 2026. It is the world's first regulatory framework built specifically for AI agents, with mandatory filing, compliance testing and product recalls in high-risk sectors.
Illinois Enacts the Toughest AI Safety Law in the US So Far
Illinois Governor JB Pritzker signed SB 315 (Artificial Intelligence Safety Measures Act) into law on July 6, 2026, described as the strictest AI safety legislation in the United States to date.
ANPD and UNDP hire consultants to study AI agents and data protection in Brazil
Brazil's National Data Protection Authority (ANPD), in partnership with the UNDP, published five calls on July 7, 2026 to hire consultants for strategic technical studies, one of them focused on AI agents in the context of personal data protection.
UN creates global AI commission with tech CEOs and heads of state
The UN and the International Telecommunication Union launched, on July 1, the AI for Good Global Commission, co-chaired by Salesforce's Marc Benioff and Rwanda's President Paul Kagame, bringing together more than 40 founding members.
What the Fable 5/Mythos 5 episode reveals about the future of AI export control
The June standoff between Anthropic and the US government shows frontier models are now treated as strategic assets — with all the unpredictability that brings for those building products on top of them.
Brazil's Chamber of Deputies again postpones vote on AI legal framework
Even with the committee report expected on June 9, Chamber Speaker Hugo Motta again postponed review of Bill 2338, conditioning progress on prior alignment with the Senate.
European Union appoints scientific panel to strengthen AI Act enforcement
The European Commission formed a Scientific Panel with 60 experts and an Advisory Forum to support enforcement of general-purpose AI models under the AI Act.
OpenAI restricts GPT-5.6 access at the request of the US government
The GPT-5.6 preview was released to only about 20 companies with prior US government approval — a new pattern for model assessment ahead of public launch.
US government lifts export controls on Fable 5 and Mythos 5
On June 30, the US Department of Commerce lifted the export controls placed on Fable 5 and Mythos 5, closing out the episode that began on June 12.
US government orders suspension of Claude Fable 5 and Mythos 5
Three days after launch, the US government ordered Anthropic to suspend foreign nationals' access to the Fable 5 and Mythos 5 models, citing national security.