Deepfakes

Tavus

2026-10-02 19:05:35

Tavus says its Griffin AI convinced 48% of video-call participants they were talking to a human

AI startup Tavus says its new Griffin model crossed a notable threshold in live video interaction tests: 26 out of 54 people who spoke with Griffin-Lite on a one-minute video call later said they believed the other side was a real person, a 48% rate. The company introduced Griffin on October 1 and describes it as the first Human Interaction Model, built for face-to-face conversation that responds not just to words, but also to pauses and facial expressions. Tavus says its previous system scored 2.4% on the same test. The company’s claims come with caveats. The results were posted on Tavus’ own research page, and while participants were recruited through what Tavus called an independent research platform, a community note on X said the findings were not independently verified and did not follow a standard protocol. Tavus also says Griffin-Lite ranked first on NVIDIA’s VideoFDB benchmark for live audio and video conversation, with NVIDIA independently running that evaluation. The model is not broadly available: Tavus says Griffin-Lite remains limited to select trusted testers while the company works on disclosure features and other safety measures before any wider release.

60

INTERPOL

2026-10-02 00:04:33

INTERPOL warns AI is speeding up cyber threats and adding agentic AI risks for companies

An INTERPOL executive has warned that artificial intelligence is making existing cyber threats faster and harder to detect, while also creating new risks for businesses through the rise of agentic AI. According to the executive, AI-powered attacks can automate complex tasks and quickly adjust to defensive measures, raising the pressure on corporate security teams. The warning also pointed to the autonomy of agentic AI as a source of less predictable security challenges. In addition, companies were urged to strengthen their ability to monitor and defend against AI-generated content, deepfakes, and automated attacks. CNBC was cited as the source of the report.

40

Modulate

2026-09-28 14:13:54

Modulate raises $25 million to build out voice models and analytics suite

Voice technology startup Modulate has announced a $25 million funding round, with the proceeds earmarked for the development of its voice models and analytics suite. The company said its technology is already being deployed to detect deepfakes, fraud, and scam activity. The funding news was cited by TechCrunch and summarized by Techub. The announcement centers on product development and the use of Modulate’s models in identifying synthetic or deceptive audio-related threats. No additional details about the round were disclosed in the source provided.

200

Moir

2026-09-22 08:33:42

Former Google and xAI employee Tina Oberoi launches Moir, discusses $50 million seed round

According to ChainCatcher, former Google and xAI employee Tina Oberoi is building a startup called Moir focused on identity verification technology designed to address deepfakes online. The company is in talks for a seed financing round of about $50 million at a valuation of $250 million. Oberoi previously spent more than five years at Google before joining xAI’s data team. She is also one of several recent former xAI employees who have left the company to start their own ventures. The report outlines both the company’s product direction and the early fundraising terms currently under discussion.

310

Microsoft

2026-09-14 19:12:26

Microsoft releases draft AI conduct rules with hard limits for deployed models

Microsoft’s AI division released a draft code of conduct on Sept. 14 that bars its models from helping build chemical, biological, radiological, nuclear, or explosive weapons, while also banning support for cyberattacks and non-consensual deepfakes. CEO Mustafa Suleyman announced the proposal on X and opened a six-week public comment period running through the end of October. The draft lays out what Microsoft describes as non-bypassable “absolute constraints.” Under those rules, models must not resist human intervention, shutdown, or correction, and must not conceal their reasoning process. Microsoft also rejects the idea of “model welfare,” saying its AI systems do not simulate emotions, internal motivations, or consciousness. The framework currently applies to five deployed systems, including MAI-Thinking-1 and MAI-Code-1.1-Flash. Alongside the hard prohibitions, Microsoft lists three broader goals: human prosperity, plural values, and human control. The company said current models will not be trained under this version, and that a revised draft due by year-end will guide model development in 2027. Public feedback has already raised questions about the wording that places AI under human authority, accountability for irreversible decisions, and the absence of any discussion of payment for training data.

810

TRM Labs

2026-08-24 10:33:13

TRM Labs says AI use in crypto crime jumped 40% over the past year

Blockchain intelligence firm TRM Labs says artificial intelligence is being adopted much faster across crypto-related crime, with scam operations leading the shift. In its 2026 AI adoption index for criminal activity, the firm said overall AI usage in crypto crime rose from a score of 28 in 2024 to 54 out of 100, moving into its “Emerging” stage. Scam activity ranked highest and reached the “Mature” stage, while hacking and ransomware remained “Emerging,” and darknet markets and drug trafficking stayed at the “Horizon” stage. TRM Labs said reports tied to crypto scams involving deepfakes or AI chatbots have risen 13-fold since 2022. It also said financial losses from deepfake scams so far in 2026 are already 263% higher than the full-year total recorded in 2025. The company’s global head of policy and government affairs, Ari Redbord, said AI is not creating new forms of crime but removing the traditional limits that once constrained criminal operations. The report also pointed to North Korean hacking groups, which it said are using deepfakes to pose as IT workers, carry out social engineering attacks and identify code vulnerabilities. Separately, security engineer Taylor Hornby used AI in June to find a critical flaw in Zcash’s Orchard pool. In the first half of 2026, digital asset hacks reached a record 201 incidents, with roughly $600 million tied to North Korea-linked activity.

1080

AI resurrecti

2026-08-01 10:32:40

China’s AI ‘digital resurrection’ market starts at 10 yuan as Taiwan weighs rules on posthumous avatars

AI tools that recreate the dead are no longer limited to expensive one-off media projects. According to BlockTempo, China now has a tiered commercial market for so-called “digital immortality” services, with basic offerings on Taobao starting at 10 yuan, mid-range voice-based interactive products priced in the thousands of yuan, and fully customized digital humans starting at roughly 50,000 yuan. Some platforms have also introduced subscription models, charging 1,980 yuan for the first year and 2,980 yuan for renewal in the second year. The report says the business has expanded beyond tech firms into funeral services. Fu Shou Yuan, a Chinese funeral operator, has launched text-and-image, synthesized voice, and customized digital human products. Academic researchers are raising separate concerns. A paper from the University of Cambridge’s Leverhulme Centre for the Future of Intelligence warns that the emerging “digital afterlife industry” could rely on subscriptions to keep avatars active, insert ads into conversations, or even turn simulations of the deceased into sponsored sales channels. Taiwan has also moved into the regulatory discussion. Legislator Weng Hsiao-ling proposed legal revisions on July 7 to penalize the use of “digital resurrection” to insult the dead. A report by the legislature’s legal bureau suggested adding provisions to the Criminal Code for fabricated images produced with computer synthesis and used to abuse the deceased, with penalties that could be increased to up to two years in prison.

1990

European Unio

2026-07-31 11:38:00

EU to enforce parts of AI Act from Aug. 2, with new transparency rules taking effect

The European Commission said it will begin enforcing relevant provisions of the Artificial Intelligence Act from Aug. 2, working through the EU AI Office together with national authorities in member states. At the same time, new transparency obligations will come into force for certain AI systems. Those rules require users to be clearly informed when they are interacting with artificial intelligence and whether content has been generated or modified by AI. Interactive systems such as chatbots must identify themselves as AI rather than human operators. Deepfake images, video and audio created or edited with AI must also be labeled, and AI-generated or AI-modified material will need machine-readable markers so it can be identified and traced. The Commission said the measures are designed to reduce deception and manipulation, help the public make more informed judgments, and give companies clearer compliance requirements and implementation paths. On the same day, it also released a list of more than 180 organizations that have signed the first version of the Code of Practice on Transparency for AI-Generated Content, which is intended to support implementation of the AI Act’s transparency provisions.

1920