deepfakes

“It’s probably not your mother.” Google has set out to end deepfakes in calls once and for all

by

Google has just given a twist to its system for detecting possible fraudulent calls, focusing on one of the most current problems related to them: identity theft using deepfakes. Through an update to the Phone app, all devices with Android 12 and above will be safer than ever starting this month. The problem. Despite the efforts of both operators and the Government itself, in Spain More than 135 million calls were blocked and five million SMS with fraudulent intentions. Globally, about 30% of these calls have fraudulent intent They were deepfakes. According to data from the March 2026 Global Financial Fraud Threat Assessment According to Interpol, the impersonation fraud led to more than 400 billion losses. How they achieve it. The scammers, as Google explains on its blogare able to spoof the phone number by routing calls through software. In other words, it is relatively easy to impersonate another person’s phone number and, when they call us, it appears as a sender that we have saved. If they manage to clone the voice using artificial intelligence, the result can be disastrous, since it is practically indistinguishable from reality. The new. Starting this month, the Google Phone app will be updated with automatic spoofing detection. It is an addition to the detection of SPAM which we already had, capable of detecting deepfakes in an advanced way. When a contact calls us and the Phone app is being used, the sending mobile phone sends a signal to our phone via RCS. So that? To verify that the call is coming from that contact’s legitimate device, and not from a scammer who has spoofed the phone number. That is to say: even if someone manages to copy a number and impersonate it, the additional authentication that Google requires when making the call cannot be replicated. How will we know. The feature will arrive globally in the Google Phone app and, in the event that a fraudulent call is detected, we will see a “it may not be X person” notice. If we decide to take it, it will be our responsibility. The curious thing is that, on the call screen, the impersonated contact will continue to appear. The main limitation of this function is that it depends on whether we use, yes or no, the Google Phone app. In other words, if they call us from iOS or using another application, this additional verification system is not required. In Xataka | Record spam calls to report: what the Spanish Data Protection Agency allows and how to do it

sexual deepfakes out, mandatory AI label and millionaire fines to the private sector

by

Spain has approved the draft Organic Law for the good use and governance of artificial intelligence. The standard adapts the legislation to framework established by the European Union with the AI ​​law and establishes a series of obligations and prohibitions based on the risk classification of these systems. The regulations impose mandatory human supervision in high-risk systems and extend responsibility beyond the company that deploys these systems, also reaching those who use them. Sanctions. In mild cases, penalties range from 6,000 euros, such as not cooperating with the authorities, to 500,000 euros or 0.5% of the total turnover. In more serious cases, turnover shoots up to 35 million or, failing that, 7% of global turnover in very serious cases. Fines of 15 million or 3% of turnover are also proposed when high-risk systems are used without human supervision. The double standard. The controversy arises because these fines are not applied when it is the administration that fails to comply with the rule. If, for example, the police or a ministry uses an AI system classified as prohibited, the law contemplates reprimands and disciplinary actions, but no fines. The Government defends that it has “raised the bar for self-demand” with transparency measures, among which is the creation of a public inventory of all AI systems used in administrative procedures. It also introduces the figure of the AI ​​delegate who will be in charge of coordinating its use. Deepfakes. The creation of deepfakes of a sexual nature and also the creation of child pornography with AI tools is prohibited. Deepfakes that do not fall into these categories (for example of a politician or a public figure) are not prohibited, but must be labeled as AI in a “clear and distinguishable” way from the time they are first shared. AI content tagging. The ministry establishes that videos and images must have a watermark in a corner of the image in which the acronym AI is clearly read. For audio generated with AI, that same seal must appear in the corresponding application, whether it is Spotify, Apple Music or another service. If you do not choose this label, it will be the audio itself that must incorporate a warning that it is generated with AI. The date to begin applying this labeling is August 2 of this year. Who will be in charge of controlling it. Supervision will fall mainly on AESIA, the Spanish AI Supervision Agency, but will be supported by other authorities such as the Spanish Data Protection Agency for biometric data, the General Council of the Judiciary for the judicial field and the Bank of Spain for everything related to the financial system. According to The CountryAESIA plans to hire 50 analysts before the end of the year to carry out this task. When it comes into effect. The law is intended to come into force before the end of 2027, but the obligation to label data with AI will come into force on August 2. However, as the law is not yet applied, in practice it means that for more than a year there will be obligations in place, but with no real ability to impose fines on companies that fail to comply. Image | Moncloa In Xataka | Deepfakes are much more than a bad joke. Now the Government wants them to be a violation of the right to honor

Deepfakes are much more than a bad joke. Now the Government wants them to be a violation of the right to honor

by

The year started with X filling up with photos of women in bikinis. Everything was normal, except that it was other users who “undressed” them using Grok, Elon Musk’s AI. In the midst of the revived debate about deepfakes, the Government has announced a new draft law with which they seek to combat them. Against deepfakes. The text It is a modification of the organic law of civil protection of the right to honor, personal and family privacy and one’s own image. According to the Minister of Inclusion, Elma Saiz, “it is a more protectionist text, adapted to new technologies.” The “ultra-impersonations carried out with artificial intelligence” or deepfakes They will be a crime when the affected person does not give their consent and the objective is to undermine their moral integrity, generating sexual or humiliating content. In these cases there may be a prison sentence of up to 2 years. The draft also raises the age of consent for image transfer to 16 years (currently it is 14 years). However, the text continues to consider use to damage the reputation of the affected person to be illegitimate, even if they have given their consent. After death. The main novelty of this reform is that it contemplates the protection of the image or voice even after the death of the person, as long as it has been specified in the will. As they point out in The Countrythis could directly affect some true-crime content in which it is used AI to recreate the image or voice of murder victims. Another case that is considered is when the perpetrator of a crime tells details of the crime in podcasts, interviews or other media. If your story reopens the victim’s wound, it will be considered an unlawful interference with their rights. Let us remember the case of book by José Bretón. The exceptions. Those that already existed in the old law are maintained, such as recordings authorized by a judge or the publication of private conversations, as long as their content is news of general interest. The novelty is that specific AI exceptions are included. The image or voice of a public figure may be used if it is in a creative or humorous context. Of course, they must clearly specify that AI has been used in its creation. Was it necessary? This is the question that some lawyers like Borja Adsuara in his X profile. His argument is that the current law already protects the right to honor in all areas, so it was not necessary to mention new technologies such as social networks, AI or deepfakes. However, it must be taken into account, as they point out in Reutersthat the European Union is requiring member countries to regulate deepfakes, especially those with non-consensual sexual content, by 2027. Previous cases. The Grok case has reignited the debate about deepfakes by the volume of images generatedbut it is not the first time that this type of practice occurs. In 2023 there was the first massive case in Spain when some teenagers generated fake nude images of several minors. Recently we also learned about the first fine from the Spanish Data Protection Agency for a minor who used an app to “undress” a classmate. Image | Unsplash (edited) In Xataka | The United Kingdom is tired of people bypassing porn blocking: their new idea is to block it on iOS and Android

Alibaba has new Open Source to generate videos. The problem is that it is being used to generate pornographic deepfakes

by

Last week Alibaba launched Wanx 2.1a new AI model for video generation that competes with others such as Sorafrom OpenAi, or I see 2from Google. The tool ended up becoming 24 hours later the person responsible for dozens of pornographic videos They will be published on the Internet. Wanx 2.1 is also an Open Source model whose code It is available in github. Very soon users fond of pornographic content and who also possessed the appropriate technical knowledge took advantage of that fair model for that: to generate porn videos. They told it in 404 averagewhere they indicated how in communities dedicated to producing and sharing pornographic deepfakes without the consent of the people represented, the users were “salivating” because of the advanced Alibaba model. Some of those videos were shared in Civitai, A platform in which users share images and videos generated with AI tools, but in which such images are also bought and sold. The statistics that we can find in the pages of the AI ​​models used to create those images and video show how These models have already been downloaded hundreds of times by users. There are already dozens of pornographic videos created with WAN 2.1, indicate in 404 average. In Civitai, users are allowed to share AI models, but unseeled pornographic content is not allowed. That does not prevent users from discharging the models and use them to produce these models themselves, they affirm in this medium. The problem is once again the use of tools that, of course, allow perfectly appropriate and spectacular videos, but can also be used for uses such as these. We already saw how this type of Deepfakes They can become very profitableand the advance of these tools makes it difficult to control this type of content. In Xataka | In South Korea, Deepfake Porn has become a nightmare. Your solution: three years in jail for those who see it

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.