Forgot your password?
Close
typodupeerror

Submission + - Should AI Ban Users Without Human Review? (medium.com)

Submitted by VTAndrew
VTAndrew writes: Artificial intelligence is rapidly becoming part of content moderation and account enforcement across major online platforms. While AI can help identify spam, scams, and harmful content at internet scale, what happens when the system gets it wrong?

A recently published Medium article examines this question through the experience of a Facebook account suspension that was reportedly initiated by an automated system, followed by an automated appeal denial and no meaningful path to human review.

The article argues that the issue isn't AI itself—it's allowing AI to become investigator, decision-maker, and appeals process without effective human oversight.

The broader concern is that platforms like Meta have evolved into critical pieces of modern infrastructure. They host community groups, school communications, local government announcements, business pages, political discussions, and years of personal history. Their ecosystems also span multiple interconnected services, meaning a single enforcement action can affect Facebook, Instagram, Messenger, Threads, and Meta hardware tied to the same account.

This concern extends beyond a single user's experience. A growing advocacy effort at People Over Platforms documents thousands of reports from users who say they were wrongfully locked out of their accounts and calls for stronger transparency, meaningful appeals, and human oversight.

The movement originated with a Change.org petition that has gathered more than 63,000 supporters before transitioning to an independent nonprofit focused on digital rights and platform accountability.

Media outlets in multiple countries have also reported on users who say they were wrongly disabled by Meta's automated enforcement systems, with some accounts later restored after additional review.

Rather than asking whether AI should be used for moderation, the article asks a different question:

If AI is empowered to make decisions that can revoke a person's digital identity, communications, communities, and purchased ecosystem, should there always be a meaningful human appeal available?

Medium article:

https://medium.com/@vtadorsett...

People Over Platforms:

https://www.peopleoverplatform...

Original Change.org petition:

https://www.change.org/p/hold-...

Submission + - When AI Becomes Judge, Jury, and Appeals Court (medium.com)

Submitted by Anonymous Coward
An anonymous reader writes: A Medium article explores the growing use of AI for account enforcement and moderation decisions on large social platforms. The author argues that the real issue isn't AI making mistakes—mistakes are inevitable—but the lack of meaningful human review when those mistakes occur.

The article describes an account suspension process in which an AI system made the initial enforcement decision, an automated appeal upheld it, and human support representatives were reportedly unable to revisit the case because it had already been marked as resolved.

The broader question raised is one of governance rather than technology: if platforms increasingly rely on AI to make decisions that can revoke access to social networks, communications, communities, purchased hardware ecosystems, and digital identities, what level of human oversight should be required?

The article also discusses the concept of "blast radius" in system design, arguing that centralized digital identity systems amplify the consequences of false positives when enforcement actions cascade across multiple services.

What level of human review should be required when AI systems are empowered to make decisions with significant real-world consequences?

Submission + - Mushroom Behind 'Tiny Human' Visions Lacks Genes For Known Psychedelics (sciencealert.com)

Submitted by alternative_right
alternative_right writes: When eaten undercooked, the mushroom can produce vivid visions of miniature people â" not unlike Gulliver on his travels to Lilliput.

"Biosynthetic gene mining of the L. asiatica genome found no close hits with any genes known in the production of mushroom psychoactive compounds," write the researchers in their published paper.

"This supports our hypothesis of the presence of a novel unidentified metabolite responsible for the unique hallucinogenic properties of L. asiatica."

Comment Printed Integral Munition Systems (Score 1) 48

by couchslug (#66204890) Attached to: The Secret Revolution in Battery Technology: 3-D Printing

Integral drone batteries would reduce complexity and likely
enhance structural strength. Warheads could similarly integrate explosive filler (with removable fuses for the usual transportation and handling safety reasons) and projectiles. Circuit boards could be rigidly supported without complex parts.

The simpler and more water and weatherproof a munition the better. Printing makes prototyping and mission-specific custom loads easy to fab. A nearly-non-metallic, low observable AI-assisted integral munitions drone package could work wonders on the battlefield and enable David to deter Goliath.

Submission + - How a Seemingly Harmless Image Can Jailbreak AI (nerds.xyz)

Submitted by BrianFagioli
BrianFagioli writes: Florida International University researchers have developed a technique called JaiLIP (Jailbreaking with Loss-guided Image Perturbation) that uses subtle image modifications to bypass AI safety guardrails. Unlike traditional jailbreaks that rely on carefully crafted prompts, the attack works through images that appear normal to human viewers.

The researchers tested the technique against BLIP-2, a multimodal AI model, and found that manipulated images significantly increased the likelihood of harmful responses. According to the study, the approach outperformed previous image-based jailbreak methods and nearly doubled the number of unsafe outputs generated during testing.

The findings highlight a potential security risk for businesses deploying AI systems that process both images and text. While most discussions about AI safety focus on prompts, the research suggests that seemingly harmless images may also serve as an attack vector.

Submission + - AI lawyer enables freelancer to win in court (theguardian.com)

Submitted by Bruce66423
Bruce66423 writes: An artificial intelligence law firm has won a case in an English court, in what is believed to be the first time a trial has been won using an AI lawyer.

A freelance HR consultant, Tamires Camal Taquidir, paid the firm, called Garfield AI, about £400 to send a legal letter and then issue court proceedings over an unpaid debt of £7,000.

Submission + - Cloudflare wants to kill the CAPTCHA and it has browser giants on board (nerds.xyz)

Submitted by BrianFagioli
BrianFagioli writes: Cloudflare has announced a new initiative with Mozilla Firefox, Google Chrome, Microsoft Edge, and Shopify to develop a privacy-focused protocol called Private Access Control Tokens (PACT). The goal is to help websites distinguish legitimate users and authorized AI agents from abusive automated traffic without relying on CAPTCHAs, invasive tracking, or browser fingerprinting.

PACT would allow trusted services to issue anonymous tokens that browsers can present to other websites as proof that a human is involved, while avoiding the disclosure of personal identity information or browsing history. The companies plan to submit the protocol for standardization.

Cloudflare argues that existing anti-bot tools are becoming less effective as AI-powered agents become more common across the web.

Submission + - ShinyHunters Published 45GB of Madison Square Garden Facial Recognition Records (thenextweb.com)

Submitted by ArchieBunker
ArchieBunker writes: The cybercrime group ShinyHunters has published 45 gigabytes of data stolen from Madison Square Garden Entertainment after the company missed a June 15 ransom deadline. The dump includes facial recognition surveillance records, internal threat assessments, and personal information from what the hackers claim are 26 million customer and corporate records. A federal class action lawsuit was filed the following day.

The breach occurred on June 5, according to a ShinyHunters spokesperson who spoke to 404 Media. The data was published on June 16, days after the New York Knicks won the NBA Finals in five games against the Spurs, putting intense public attention on the arena and its owner, James Dolan.

What makes this breach unusual is the nature of the surveillance data it exposed. MSG has deployed facial recognition technology across its venues for years, using the system to screen visitors and, controversially, to ban lawyers from firms that have sued the company. The leaked files include biometric tracking logs, background check information, internal threat assessments, and what the class action complaint describes as detailed dossiers on attendees.

A sample reviewed by 404 Media contained files specifically referencing Knicks-related personalities, with fields including “address,” “claim to fame,” “cost of talent,” and direct contact information for individuals or their representatives. The data also included internal risk tags classifying celebrities: actor Ben Stiller was profiled as “low risk,” while rapper A Boogie wit da Hoodie was flagged as “high risk,” according to the class action filing. No documented criteria explaining the labels were included in the leaked files.

Customer emails were also part of the dump, including messages from fans who had expressed concern about being misidentified by MSG’s facial recognition cameras. The inclusion of this correspondence reveals that MSG was collecting and storing complaints about its own surveillance practices alongside the biometric data itself.

A class action lawsuit, Avalo v MSG Entertainment, was filed on June 16 in New York federal court. The plaintiff, Carlos Avalo, attended a concert at MSG in September 2025 and alleges his biometric data was captured by the venue’s entry systems. The lawsuit seeks at least $5 million in initial damages.

The complaint accuses MSG of corporate negligence in failing to secure the data it aggressively collects, despite clear warnings from privacy advocates and a previous breach.

This is MSG’s second major breach in under a year. In a separate incident disclosed in February 2026, the Cl0p ransomware group exploited a vulnerability in a vendor-hosted Oracle eBusiness Suite application used by MSG for payroll and human resources. That intrusion began in August 2025 but went undetected until December 16, 2025, and exposed the names, addresses, and Social Security numbers of roughly 131,070 individuals, primarily employees and contractors.

ShinyHunters has been on a sustained campaign in 2026, exploiting an unpatched Oracle PeopleSoft zero-day to breach more than 100 organisations, two-thirds of them universities. The group previously orchestrated the 2024 Snowflake supply chain attacks that compromised Ticketmaster and AT&T, and in March 2026 breached the European Commission, leaking 350 gigabytes of data from 42 internal clients.

The MSG attack follows the same playbook ShinyHunters used against Instructure’s Canvas learning management system in April, where the group claimed 3.65 terabytes of data from 275 million users across 9,000 schools. The pattern is consistent: identify a target sitting on large volumes of sensitive data, exfiltrate it, set a ransom deadline, and publish when the deadline passes.

MSG Entertainment has not publicly confirmed the scope of the breach or commented on the class action. The company’s facial recognition programme has faced scrutiny since at least 2022, when it drew attention for using the technology to bar attorneys from firms involved in litigation against the company. The New York attorney general investigated, and a state court initially ruled the policy violated anti-discrimination law, though an appeals court later reversed that decision.

The breach raises a question that extends beyond MSG: organisations that invest heavily in surveillance technology to monitor their visitors are creating precisely the kind of high-value data troves that groups like ShinyHunters target. The 26 million figure cited by the hackers has not been independently verified, and the full scope of the exposed biometric data remains unclear as the investigation continues.

Submission + - Russian Satellites Cosmos 2546 Have Been Jamming GPS Signals Across Europe (arstechnica.com)

Submitted by tomatocat
tomatocat writes: In 2024, Dana A. Goward, founder of the Resilient Navigation and Timing Foundation, received a call from an anonymous British researcher, He said that interference from space was more than a possibility — he had observed it. Examining data from terrestrial reference stations operated by the International Global Navigation Satellite System (GNSS) Service, he had noticed instances in which GPS signal strength had decreased markedly. In each case it was for less than ten seconds, but the events had been recorded by stations across a very broad section of northern Europe. The researcher consented to the Foundation sharing these findings. Todd Humphreys of the University of Texas at Austin and his student Zach Clements analyzed ground station data spanning from January 2019 to April 2026; they identified 75 days with at least one widespread GNSS interference event. The paper mentioned (PDF), "The interference peak is centered at 1577.5 MHz, about 2 MHz above the GPS L1 center frequency of 1575.42 MHz. In addition to tracked GPS L1 C/A signals, tracked Galileo E1 and BeiDou B1C/B1A signals also exhibited a concurrent drop in CNR during interference events." Humphreys and his colleagues calculated that the source had to be at least 1,200 kilometers above the Earth, But they couldn’t go further. Later, Humphreys received an email stating that radio stations in Amsterdam, Netherlands, and Trondheim, Norway, had captured raw interference signal data on February 11, 2026. By examining the difference in timing when that signal arrived at the two different stations, Humphreys and Clements calculated a “quasi-hyperboloid surface”, stretching tens of thousands of kilometers into space where the interference satellite must have been located. The margin of error represented by the thickness of that surface was only five meters. A comparison of suspect satellite orbits with the quasi-hyperboloid surface showed that only one satellite’s orbit aligned perfectly—the Russian satellite Cosmos 2546, which are designed to provide early warnings when they detect ballistic missile launches. The research paper is published at https://arxiv.org/abs/2606.03673. This discovery has raised concerns regarding Russian electronic warfare capabilities. An EU spokesperson told The New York Times that the EU has launched an investigation into these incidents but that the results remain classified, while The press office for the Russian Embassy in Washington, D.C. said they don't have a comment on that.

Submission + - Linear A Minoan script maybe deciphered by linguistics researcher with AI help (aiclambake.com) 1

Submitted by Aristos Mazer
Aristos Mazer writes: A researcher claims to have deciphered the millennia-old Linear A script. After studying for 7 years, Tom Di Mino had an insight about the structure of one sample of Linear A that lead Tom to compare it against an ancestor of Hebrew. Tom created an AI analysis that yielded full translations of Linear A, and correlated against the known Linear B for verification.

Comment Defending idiots from themselves isn't the .govs (Score 1) 57

Slashdot doesn't need this clickbait.

A stupid rich vain asshole killed people so intensely silly they cared about Titanic, whose sole claim to fame IS fame. The world is slightly wiser in consequence.

Think about it. There is no reason a functioning adult should be morbidly fascinated by a mere shipwreck but people crave to masturbate to drama, and romantic death appeals to the bitch-made (a perfect hood term for a much wider degeneracy) mind.

The other casualties were so cravenly silly they utterly failed to perform THEIR OWN due diligence before becoming someone else's suicidal beta testers. They were sufficiently educated to understand the basics, including that ZERO reason exists not to copy proven hull designs zero reason existed to change let alone use an utterly absurd choice of hull.

Rush had Alvin--tier money but pure vanity is why he chose a childishly silly hull design no reason existed to want because nothing about it was better. His loss is as minor as a common auto accident. Ditto the "collateral damage" who knew what they signed up for.

Submission + - Microsoft discovers new lightweight backdoor that steals cryptocurrency (arstechnica.com)

Submitted by joshuark
joshuark writes: Ars Technica reports Microsoft says it has detected new self-propagating malware that spreads through USB drives in search of cryptocurrency credentials, which it then sends to attacker-controlled servers. The company named the worm Crypto Clipper because it monitors the contents of device clipboards for patterns consistent with wallet addresses or seed phrases.

“The execution of this clipper is notable because it does not depend on a traditional installer or exposed IP-based C2 infrastructure,” Microsoft said Thursday. “Instead, it deploys a portable Tor client, routes traffic through a local SOCKS5 proxy, and blends data theft with remote code execution, turning a financially motivated stealer into a lightweight backdoor.”

“This malware family shows how lightweight, script-based stealers can deliver outsized impact when paired with anonymized communications and runtime tasking,” Microsoft said. “The combination of Tor-routed C2, clipboard targeting, screenshot capture, and remote code execution gives attackers both immediate monetization paths and continued control over compromised devices.”

Big question is "What's in your crypto wallet?"

Submission + - OpenAI just exposed how bad AI still is at real science (nerds.xyz)

Submitted by BrianFagioli
BrianFagioli writes: OpenAI introduced LifeSciBench, a new benchmark designed to evaluate AI systems on realistic life science research tasks rather than simple biology questions. While OpenAIâ(TM)s top-performing GPT-Rosalind model led the rankings, it achieved a pass rate of just 36.1 percent, failing nearly two-thirds of benchmark tasks. The company says the results highlight progress in scientific communication and evidence synthesis, but also reveal persistent weaknesses in artifact-heavy and design-oriented scientific work.

Submission + - SMPTE Opens Entire Standards Library to Public at No Cost (smpte.org)

Submitted by innocent_white_lamb
innocent_white_lamb writes: "SMPTE®, the home of media professionals, technologists and engineers, has announced that its entire Standards catalog is now freely available to the global media technology community. This includes all published SMPTE Standards, Recommended Practices, Engineering Guidelines and Registered Disclosure Documents (RDDs), as well as all future releases. For more than a century, SMPTE Standards have helped enable the interoperability that underpins the entertainment technology industry. By removing barriers to access, this milestone is expected to accelerate adoption and implementation, strengthen interoperability, and help drive the next generation of innovation."

SMPTE is the Society of Motion Picture and Television Engineers, a global professional organization that sets standards for film, television, and digital media

Slashdot Top Deals

Real computer scientists don't comment their code. The identifiers are so long they can't afford the disk space.

Close