March 2024 - nexa - server-nexa.polito.it

Fwd: [mpiwg-announce] AI and Digital seminar series
by J.C. DE MARTIN March 16, 2024

March 16, 2024

> > *From: *Etienne Benson <ebenson(a)mpiwg-berlin.mpg.de> > *Subject: **[mpiwg-announce] Fwd: AI and Digital seminar series* > *Date: *March 15, 2024 at 5:11:43 PM GMT+1 > *To: *Mpiwg Announce <mpiwg-announce(a)mpiwg-berlin.mpg.de> > >> AI and the Digital - Final Poster PNG copy.jpeg

1 0

The Software Heritage Archive wants to deadname me forever: part 1
by Giacomo Tesio March 15, 2024

March 15, 2024

When I first contacted them, politely, they didn't even respond to me. What happened instead was that their content policy changed from a standard EU content policy to one that is very defensive about their ability to archive personal data forever with no exceptions. [...] The reason they can't just change references to your deadname, like any decent person would, is vaguely because of "integrity" (the common refrain of transphobic cowards), but more specifically because they made a fucking immutable blockchain out of it. [...] You can guess what happened next: I fucking exploded at them on Twitter. This got a faster reaction than anything else I'd seen in the process. The rant against changing names in the content policy disappeared again. One of the developers went to my DMs to tell me condescendingly that I'd understood it wrong. It wasn't a blockchain, he explained, it's just a Merkle tree, like Git, because we copied your Git repository. I screenshotted their documentation that said "blockchain". [...] He said, but we have to do that, what if someone wanted to cite your code in a paper? You can't just change a citation. I pointed out that you can just change a citation if it's wrong, and that I've gotten citations that deadname me changed. I pointed out that people cite my code under my correct name through Zenodo, not through whatever cryptographic bullshit SWH was dreaming up. He said, well anyway, I don't see why this is our problem. You can't even change your name in your own repository! Imagine if people could just change history ha ha I pointed him at my Git repository that I had already mutated using git-filter-repo. He stopped replying. Continua su https://cohost.org/arborelia/post/4968198-the-software-heritag Questa storia presenta molti aspetti politicamente interessanti. Anzitutto evidenzia come l'ignoranza tecnica causi alienazione cibernetica dove meno ce la si aspetta: modificare un repository git è facilissimo, SH avrebbe potuto riclonarlo in pochi minuti da github già aggiornato e con qualche giorno di lavoro avrebbe anche potuto introdurre dei redirect dalle url con i vecchi hash a quelle con i nuovi. MA, non sapendolo, hanno tirato su questo ridicolo muro di gomma, trattando una persona che esercita i propri diritti come un problema di conformità alla macchina: in sintesi credendo che la macchina non si possa adattare ai diritti di una persona, si chiede a quella persona di rinunciare ai propri diritti. Non è la prima e non sarà l'ultima volta, ma questa volta al posto della mistica "AI" troviamo semplici repository GIT. E poi interessantissimo come tale ignoranza/alienazione venga scambiata per transfobia dall'autrice che definisce "transphobic" "white cis" i dirigenti della Software Heritage. Un'accusa ovviamente infondata, ma che non può non richiamare alla memoria un'accusa altrettanto strumentale sottoscritta da alcuni di quei dirigenti contro un'altra vittima [1]. In quell'occasione però, misogynist, ableist e transphobic era una singola persona, definita come "forza pericolosa" e indicata con nome e cognome. In secondo luogo porta alla luce come, anche nella civilissima Francia, i DPO dimentichino spesso lo scopo del proprio ruolo: garantire la protezione dei dati personali, non degli interessi dei propri clienti. Fortunatamente in questo caso, la caparbietà del richiedente potrebbe far coincidere la negligenza del DPO (che avrebbe dovuto dar seguito alla richiesta di esercizio dei diritti, non riscrivere la privacy policy sperando di esentare la Software Heritage) con un meritatissimo danno di immagine al proprio cliente. In terzo luogo, è interessantissimo come l'esercizio dei diritti sia dovuto passare da Twitter. È interessante non solo perché rende evidente come chi ha partecipato o scatenato gogne come quella subita anni fa da Stallman, le teme molto di più di quanto non tema un Garante serio come quello francese. E d'altro canto, si tratta di un timore infondato: Twitter non ha alcuna ragione di catalizzare uno shitstorm contro una organizzazione finanziata, fra gli altri da Google, Microsoft, Intel e molti altre aziende con interessi strategici affini [2]. Stallman invece può davvero sembrare "una forza pericolosa nel software libero" per quegli interessi. [3] Infine sarà interessante vedere come andrà a finire (sembra che l'autrice si stia preparando a dare battaglia sul piano legale). Personalmente scommetto che finirà tutto a tarallucci e vino: un bel blog di scuse con tanto di ringraziamento nei confronti dell'autrice e di auto-assoluzione, la sostituzione del nome, una vittoria giustamente sbandierata dalla stessa autrice (e probabilmente strumentalizzata da qualche movimento). Alla fine, questa volta, il diritto verrà esercitato con successo. Speriamo ne nasca un precedente utile per tutti! Giacomo [1] https://rms-open-letter.github.io/ [2] https://www.softwareheritage.org/support/sponsors/ [3] che lo sia ancora, in realtà, non saprei dirlo...

4 15

AI models with systemic risk
by Antonio March 15, 2024

March 15, 2024

Articolo 52 dell'AI Act [1]. Classification of general-purpose AI models as general-purpose AI models with systemic risk ... 2. A general-purpose AI model shall be presumed to have high impact capabilities pursuant to paragraph 1, point (a), when the cumulative amount of computation used for its training measured in FLOPs is greater than 10^25 Dato che quel 10^25 è il discrimine tra AI a rischio sistemico e AI non a rischio sistemico, bisogna essere chiari. Cosa intendeva il legislatore per FLOPs ? Plurale di FLOP, ovvero, QUANTITA' di operazioni in virgola mobile, oppure FLOP al secondo, ovvero, operazioni in virgola mobile AL SECONDO ? In entrambi i casi è sbagliato. Nel primo, è un errore "morfologico", il plurale di FLOP è FLOPS, non FLOPs. Nel secondo, ovvero nel caso il significato fosse di FLOP/s, in questo caso è un errore "matematico" di stima, perché 10^25 è milioni di volte superiore alla "potenza" dei vari GPT4, Gemini, ecc. FLOPs vs FLOPS When dealing about computing effort and computing speed (hardware performance), terminology is usually confusing. The term ‘compute’ is usually ambiguous, sometimes applied for a number of operations or the number of operations per second. However, it is important to clarify what kind of operations and the acronyms for them. In this regard, we will use the acronym FLOPS to measure hardware performance, by referring to the number of floating point operations per second, as standardised in the industry, while FLOPs will be applied to the amount of computation for a given task (e.g., a prediction or inference pass), by referring to the number of operations, counting a multiply-add operation pair as two operations. For instance, we found out that the acronym FLOP may be misleading. By FLOP, we mean one floating point operation, a measure of the amount of compute (computing effort) and by FLOPS, we mean floating point operations per second, i.e., FLOPS = FLOP/s. However, many papers, especially CV papers, use the terms FLOPs and FLOPS to refer to the number of operations, but we will be just use FLOPs as the plural of FLOP, never as FLOPS. Then there is the question of what a FLOP is. When dealing with DNN, this is usually associated with the number of multiply-add operations, even there are other type of operations involved when executing a DNN. This is done this way because it is usually a good estimation [Hollemans, 2018, Clark et al., 2020]. More specifically, we will count one fused multiply-add operation as 2 FLOPs (note the lowercase ‘s’). Hardware manufacturers count them in this manner [NVIDIA, 2015], because in fact there are two mathematical operations. However, CV research papers count a multiply-add operation as only one operation. In this case, we will multiply the number of operations reported by 2. In sum, the acronym FLOPS will be applied to measure hardware performance, by referring to the number of floating point operations per second, as standardised in the industry, while FLOPs will be applied to the amount of computation for a given task (e.g., a prediction or inference pass), by referring to the number of operations, counting a multiply-add operation pair as two operations." [2] A. [1] https://www.europarl.europa.eu/doceo/document/TA-9-2024-0138_EN.pdf [2] https://arxiv.org/pdf/2109.05472.pdf

1 0

How Facebook Messenger and Meta Pay are used to buy child sexual abuse material | Technology | The Guardian
by Alberto Cammozzo March 15, 2024

March 15, 2024

<https://www.theguardian.com/global-development/2024/mar/14/facebook-messeng…> How Facebook Messenger and Meta Pay are used to buy child sexual abuse material When police in Pennsylvania arrested 29-year-old Jennifer Louise Whelan in November 2022, they charged her with dozens of counts of serious crimes, including sex trafficking and indecent assault of three young children. One month earlier, police said they had discovered Whelan was using three children as young as six, all in her care, to produce child sex abuse material. She was allegedly selling and sending videos and photos to a customer over Facebook Messenger. She pleaded not guilty. The alleged buyer, Brandon Warren, was indicted by a grand jury in February 2022 and charged with nine counts of distribution of material depicting minors engaged in sexually explicit conduct. Warren also pleaded not guilty. Court documents seen by the Guardian quote Facebook messages between the two in which Warren allegedly describes to Whelan how he wants her to make these videos. “I’ll throw in a little extra if you tell him it makes mommy feel good and get a good length video,” he tells Whelan, according to the criminal complaint document used for her arrest. Whelan received payment for the footage over Meta Pay, Meta’s payment system, according to the criminal complaint against him. “Another 250 right? Heehee,” she allegedly wrote to Warren after sending him a video of her abusing a young girl. Meta Pay, known as Facebook Pay before rebranding in 2022, is a peer-to-peer payment service enabling users to transfer money over the company’s social networks. Users upload their credit cards, debit cards or PayPal account information to Facebook Messenger or Instagram to send and receive money. A spokesperson for Meta confirmed that the company has seen and reported payments via Meta Pay on Facebook Messenger that are suspected of being linked to child sexual exploitation. “Child sexual exploitation is a horrific crime. We support law enforcement in its efforts to prosecute these criminals and invest in the best tools and expert teams to detect and respond to suspicious activity. Meta reports all apparent child sexual exploitation to NCMEC [the National Center of Missing and Exploited Children], including cases involving payment transactions,” the spokesperson said. Through reviewing documents and interviewing former Meta content moderators, a Guardian investigation has found that payments for child sexual abuse content taking place on Meta Pay are probably going undetected, and unreported, by the company. Court documents show Whelan and Warren’s actions were not spotted or flagged by Meta. Instead, Kik Messenger, another social platform, reported Warren had uploaded videos suspected to be child sexual abuse material (CSAM) to share with other users. This triggered a police investigation in West Virginia, where Warren lives. His electronics were seized, and police then discovered the eight videos and five images that he had allegedly bought from Whelan over Facebook Messenger. “We responded to valid legal process,” said a Meta spokesperson, in response to the Guardian’s findings that the company did not detect these crimes. Additionally, two former Meta content moderators, employed between 2019 and 2022, told the Guardian that they saw suspicious transactions taking place via Meta Pay that they believed to be related to child sex trafficking, yet they were unable to communicate with Meta Pay compliance teams to flag these payments. “It felt like [Meta Pay] was an easy-to-use payment method since these people were communicating on Messenger. The amounts sent could be hundreds of dollars at a time,” says one former moderator, who spoke under the condition of anonymity because they had to sign a non-disclosure agreement as a condition for employment. The moderator, employed for four years until mid-2022 by Accenture, a Meta contractor, reviewed interactions between adults and children over Facebook Messenger for inappropriate content. Payments for sex or CSAM are typically just a few hundred dollars or less in cases reviewed by the Guardian. According to the former Meta compliance analyst, transactions of such small amounts are unlikely to be flagged for review by Meta’s systems. This means that payments connected to illicit activities are probably taking place undetected, financial crimes experts said. A Meta spokesperson said that the company uses a combination of automated and human review to detect suspicious financial activity in payment transactions in Messenger. “The size of the payment is just one signal our teams use to identify potentially suspicious activity, and our compliance analysts are trained to assess a variety of signals,” said the Meta spokesperson. “If our teams had reason to suspect suspicious activity, especially activity involving a child and even if the payments are small, it would be investigated and reported appropriately.” The spokesperson also said that the company had “a strong ‘see something, say something’ culture”. For situations where American men were targeting underage girls abroad to groom, payments could be for things like getting a phone and school supplies, the moderator said. “Most of what we saw were older men from America, targeting girls in Asian countries and often travelling there,” the moderator added. “When it comes to child exploitation and CSAM, it’s really all about small amounts,” said Silvija Krupena, director of the financial intelligence unit at RedCompass Labs, a London-based financial consultancy. “It’s a global crime and criminals, with different types of offenders. In low-income countries like the Philippines, $20 is big money. The production usually happens in those countries. These are small amounts that can fall through the cracks when it comes to traditional money-laundering controls.” Meta has a team of about 15,000 moderators and compliance analysts who are tasked with monitoring its platforms for harmful and illegal content. Possible criminal behavior is supposed to be escalated by Meta and reported to law enforcement. Anti-money laundering regulations also require money service businesses to train their compliance staff to have access to enough information to be able to detect when illegal financing occurs. Yet contractors monitoring Meta Pay transaction activity do not receive specific training for detecting and reporting money flows that could be related to human trafficking, including the language, codewords and slang that traffickers typically use, a former Meta Pay payment compliance analyst contractor said. “If a human trafficker is using a codeword for selling girls, we didn’t get into that. We didn’t really get trained on those,” said the former compliance analyst. “You don’t even give it a second thought or even dig into that kind of stuff at all.” A Meta spokesperson disputed the payment compliance analyst’s claims. “Compliance analysts receive both initial and ongoing training on how to detect potentially suspicious activity – which includes signs of possible human trafficking and child sexual exploitation. Our program is regularly updated to reflect the latest guidance from financial crime regulators and safety experts,” the spokesperson said. Meta’s history with accusations of child exploitation Meta’s platforms have been linked to alleged child exploitation and the distribution of CSAM in the past. In December, the New Mexico attorney general’s office filed a lawsuit against the company, alleging Facebook and Instagram are “breeding grounds” for predators targeting children for human trafficking, grooming and solicitation. The suit followed an April 2023 Guardian investigation, which revealed how child traffickers were using Meta’s platforms to buy and sell children into sexual exploitation. As a money services business, Meta Pay is subject to the US anti-money laundering and “know your client” (KYC) banking regulations, which require businesses to report illicit financing to the US treasury department’s Financial Crimes Enforcement Network (FinCEN). If Meta fails to detect and report these payments, it could be in violation of US anti-money laundering laws, financial crimes experts have said. “Regulations apply to any company that participates in a payments business. But for social media because they can see users, they see their lives, their transactions, they can see abuse and see contact. It’s such a low-hanging fruit for them to detect this,” said Krupena. Other peer-to-peer payment apps have faced scrutiny for their practices in preventing illicit activity. In 2023, Senate Democrats requested detailed fraud detection and prevention methods from PayPal, Venmo and Cash App. Sex trafficking “ran rampant” on Cash App, according to a report last year by US investment research firm Hindenburg. Block, Cash App’s owner, disputed these claims, threatening legal action. Meta introduced end-to-end encryption to Facebook Messenger in late 2023, but even before this, payment compliance analyst contractors could not access the Messenger chat between the two users exchanging funds. The former Meta compliance analyst told the Guardian their team could only see transactions with notes and the relationship between the two users. “I don’t know how you do compliance in general without being able to see intentions around transacting,” said Frances Haugen, a former Facebook employee turned whistleblower, who released tens of thousands of damaging documents about its inner workings in 2021. “If the platforms actually wanted to keep these kids safe, they could.” Siloed work prevents flagging suspicious transactions, say ex-moderators Other former content moderators interviewed by the Guardian compared their jobs to call center or factory work. Their jobs entailed reviewing content flagged as suspicious by users and artificial intelligence software and making quick decisions on whether to ignore, remove or escalate the content to Meta through a software program. They say they could not communicate with the Meta Pay compliance analysts about suspicious transactions they witnessed. “We were not allowed to contact Facebook employees or other teams,” one former moderator said. “Our managers didn’t tell us why this was.” Gretchen Peters, who is the executive director of the Alliance to Counter Crime Online, has documented the sale of narcotics, including fentanyl, over Meta’s platforms. She also interviewed Meta moderators who were not permitted to communicate with other teams in the company. She said this siloing was a “major violation” of “know your customer” banking regulations. “We’ve heard from moderators at Meta they can see illegal conduct is occurring and that there are concurrent transactions through Meta Pay, but they have no way of communicating what they are seeing internally to moderators at Meta Pay,” said Peters. A Meta spokesperson said the company prohibits the sale or purchasing of narcotics on its platforms and removes that content when it finds it. “Meta complies with all applicable US anti-money laundering laws,” the spokesperson said. “It is also untrue to suggest that there is a lack of communication between teams. Content moderators are trained to escalate to a specific point of contact, who brings in the appropriate specialist team.” In December, Meta announced it had rolled out end-to-end encryption for messages sent on Facebook and via Messenger. Encryption hides the contents of messages from anyone but the sender and intended recipient by converting text and images into unreadable cyphers that are unscrambled on receipt. Yet this move could also affect the company’s ability to prevent illicit transactions on Meta Pay. Child safety experts, policymakers, parents and law enforcement criticized the move, arguing encryption obstructs efforts to rescue child sex trafficking victims and the prosecution of predators. “When Meta Pay is linked to Messenger or Instagram, the messages associated with payments could uncover illicit behaviors,” said Krupena. “Now that this context is removed, the implications are significant. It almost feels like encryption is inadvertently facilitating illicit activity. This opens many opportunities for criminals to hide in plain sight.” A Meta spokesperson said the decision to move to encryption was to “provide people with privacy”, and that the company encourages users to self-report private messages related to child exploitation to the company. “Moving to an encrypted messaging environment does not mean we will sacrifice safety, and we have developed over 30 safety tools, all of which work in encrypted messaging,” said the spokesperson. “We’ve now made our reporting tools easier to find, reduced the number of steps to report and started encouraging teens to report at relevant moments.” FinCEN declined to comment. PayPal did not respond to a request for comment.

1 0

Identifying software (and software bill of materials).
by 380° March 14, 2024

March 14, 2024

Buonasera, segnalo un articolo un po' tecnico che illustra cosa significa identificare univocamente i "software artifact", specialmente quando in forma binaria, che è la forma particolarmente problematica del software... da sempre :-) «Identifying software» Ludovic Courtès, Maxim Cournoyer, Jan Nieuwenhuizen, Simon Tournier — March 4, 2024 https://guix.gnu.org/en/blog/2024/identifying-software/ --8<---------------cut here---------------start------------->8--- [...] 1 On Software Identification ════════════════════════════ The /Software Identification Ecosystem Option Analysis/ white paper released by CISA in October 2023 studies options towards the definition of /a software identification ecosystem that can be used across the complete, global software space for all key cybersecurity use cases/. Our experience lies in the design and development of [GNU Guix], a package manager, software deployment tool, and GNU/Linux distribution, which emphasizes three key elements: *reproducibility, provenance tracking, and auditability*. We explain in the following sections our approach and how it relates to the goal stated in the aforementioned white paper. Guix produces binary artifacts of varying complexity from source code: package binaries, application bundles (container images to be consumed by Docker and related tools), system installations, system bundles (container and virtual machine images). All these artifacts qualify as “software” and so does source code. Some of this “software” comes from well-identified upstream packages, sometimes with modifications added downstream by packagers (patches); binary artifacts themselves are the byproduct of a build process where the package manager uses /other/ binary artifacts it previously built (compilers, libraries, etc.) along with more source code (the package definition) to build them. How can one identify “software” in that sense? Software is dual: it exists in /source/ form and in /binary/, machine-executable form. The latter is the outcome of a complex computational process taking source code and intermediary binaries as input. Our thesis can be summarized as follows: *We consider that the requirements for source code identifiers differ from the requirements to identify binary artifacts.* Our view, embodied in GNU Guix, is that: 1. *Source code* can be identified in an unambiguous and distributed fashion through /inherent identifiers/ such as cryptographic hashes. 2. *Binary artifacts*, instead, need to be the byproduct of a /comprehensive and verifiable build process itself available as source code/. In the next sections, to clarify the context of this statement, we show how Guix identifies source code, how it defines the /source-to-binary/ path and ensures its verifiability, and how it provides provenance tracking. [GNU Guix] <https://guix.gnu.org> [...] As with Nix, build processes are identified by /derivations/, which are low-level, content-addressed build instructions; derivations may refer to other derivations and to source code. For instance, `/gnu/store/c9fqrmabz5nrm2arqqg4ha8jzmv0kc2f-gcc-11.3.0.drv' uniquely identifies the derivation to build a specific variant of version 11.3.0 of the GNU Compiler Collection (GCC). Changing the package definition—patches being applied, build flags, set of dependencies—, or similarly changing one of the packages it depends on, leads to a different derivation (more information can be found in [Eelco Dolstra's PhD thesis]). Derivations form a graph that *captures the entirety of the build processes leading to a binary artifact*. In contrast, mere package name/version pairs such as `gcc 11.3.0' fail to capture the breadth and depth elements that lead to a binary artifact. This is a shortcoming of systems such as the *Common Platform Enumeration* (CPE) standard: it fails to express whether a vulnerability that applies to `gcc 11.3.0' applies to it regardless of how it was built, patched, and configured, or whether certain conditions are required. [reproducible builds] <https://reproducible-builds.org> [Nix package manager] <https://nixos.org> [Eelco Dolstra's PhD thesis] <https://edolstra.github.io/pubs/phd-thesis.pdf> [...] 5 Provenance Tracking ═════════════════════ We define provenance tracking as the ability *to map a binary artifact back to its complete corresponding source*. Provenance tracking is necessary to allow the recipient of a binary artifact to access the corresponding source code and to verify the source/binary correspondence if they wish to do so. [...] In other words, because Guix itself defines how artifacts are built, *the revision of the Guix source coupled with the package name unambiguously identify the package's binary artifact*. As scientists, we build on this property to achieve reproducible research workflows, as explained in this [2022 article in /Nature Scientific Data/]; as engineers, we value this property to analyze the systems we are running and determine which known vulnerabilities and bugs apply. Again, a software bill of materials (SBOM) written as a mere list of package name/version pairs would fail to capture as much information. The *Artifact Dependency Graph (ADG) of [OmniBOR]*, while less ambiguous, falls short in two ways: it is too fine-grained for typical cybersecurity applications (at the level of individual source files), and it only captures the alleged source/binary correspondence of individual files but not the process to go from source to binary. [`guix pack'] <https://guix.gnu.org/manual/en/html_node/Invoking-guix-pack.html> [the `time-machine' command] <https://guix.gnu.org/manual/en/html_node/Invoking-guix-time_002dmachine.html> [2022 article in /Nature Scientific Data/] <https://doi.org/10.1038/s41597-022-01720-9> [OmniBOR] <https://omnibor.io/> [...] --8<---------------cut here---------------end--------------->8--- -- 380° (Giovanni Biscuolo public alter ego) «Noi, incompetenti come siamo, non abbiamo alcun titolo per suggerire alcunché» Disinformation flourishes because many people care deeply about injustice but very few check the facts. Ask me about <https://stallmansupport.org>.

1 0

ChatGPT is bad at following copyright law, researchers say
by J.C. DE MARTIN March 14, 2024

March 14, 2024

*ChatGPT is bad at following copyright law, researchers say* /Patronus AI, which evaluates AI models for mistakes, said chatbots generated copyrighted content at a "high rate"/ Britney Nguyen Updated March 6, 2024 https://qz.com/openai-chatgpt-anthropic-claude-copyright-law-violation-1851…

1 0

Pirates: AI Act fails to protect citizens’ rights – Patrick Breyer
by J.C. DE MARTIN March 13, 2024

March 13, 2024

https://www.patrick-breyer.de/en/pirates-ai-act-fails-to-protect-citizens-r… (messaggio spedito in movimento - scusate brevità ed eventuali refusi)

1 0

AI Nationalism(s): Global Industrial Policy Approaches to AI
by J.C. DE MARTIN March 13, 2024

March 13, 2024

https://ainowinstitute.org/ai-nationalisms (messaggio spedito in movimento - scusate brevità ed eventuali refusi)

1 0

The EU AI Act: a failure for human rights, a victory for industry and law enforcement
by Daniela Tafani March 13, 2024

March 13, 2024

The EU AI Act: a failure for human rights, a victory for industry and law enforcement PUBLISHED: 13 March 2024 The EU’s Artificial Intelligence (AI) Act was voted through and adopted today, 13 March, in a plenary session of the European Parliament. Despite its intention to protect people’s rights, the final text of the AI Act is an unambitious piece of product safety legislation. Following outrageous lobbying by industry and law enforcement agencies, the final text is full of loopholes, carve-outs, and exceptions, which mean that it will not protect people, nor their human rights, from some of the most dangerous uses of AI. First proposed in 2021, the EU’s co-legislative bodies finally announced a deal on the EU AI Act in December 2023. The European Parliament entered the last trilogue negotiations with a very strong text, but it fell at the final hurdle; protections for people’s rights have either been watered down or fully removed in the text voted through today. "EU officials boast about being global trendsetters when it comes to regulating AI, but with this law they’ve set the lowest bar possible. The new AI Act is littered with concessions to industry lobbying, exemptions for the most dangerous uses of AI by law enforcement and migration authorities, and prohibitions so full of loopholes that they don’t actually ban some of the most dangerous uses of AI. " Daniel Leufer, Senior Policy Analyst at Access Now Access Now notes in particular that the final version of the AI Act: Fails to properly ban some of the most dangerous uses of AI, including systems that enable biometric mass surveillance and predictive policing systems; Creates a glaring loophole via Article 6(3) for developers to exempt themselves from obligations for high-risk AI systems,; Exempts law enforcement and migration authorities from important public transparency requirements when they use high-risk AI, meaning they can continue deploying dangerous systems in secret; Further broadens the national security exemption beyond what is allowed for in the EU treaties, allowing governments to exempt themselves from obligations under the AI Act in order to pursue cases deemed relevant to national security; Creates a separate regime for people migrating, seeking refuge, and/or living undocumented, leaving with them far fewer rights than EU citizens and almost no access to remedy when these rights are violated. The #ProtectNotSurveil coalition, which Access Now co-leads, has detailed how the final AI Act presents heightened risks in a migration context, setting dangerous precedents for the use of surveillance technologies against people on the move and racialised communities. "The EU AI Act will only entrench police and migration authorities’ existing culture of impunity – encouraging them to keep deploying harmful AI systems against the most marginalised communities. Once again, the EU’s digital policies are being used as a testbed for oppressive surveillance representing a blatant attack on everyone’s fundamental rights." Caterina Rodelli, EU Policy Analyst at Access Now Despite the AI Act’s profound shortcomings, Access Now will work with civil society partners to push for the strongest possible implementation of the AI Act and the few rights-protecting tools it includes, while continuing to monitor and call out AI-facilitated abuses. https://www.accessnow.org/press-release/ai-act-failure-for-human-rights-vic…

1 0

Klarna AI assistant handles two-thirds of customer service chats in its first month
by don Luca Peyron March 13, 2024

March 13, 2024

In soldoni... nonsanno cosa farsene di 700 persone... Buona lettura dl https://www.klarna.com/international/press/klarna-ai-assistant-handles-two-… _________________________ don Luca Peyron Pastorale Universitaria - Apostolato Digitale Arcidiocesi di Torino www.universitari.to.it via XX settembre 83, Torino tel. 011 5156239

3 2