Search: “Decoded Legal”

Lawyer fined $5K over AI-hallucinated witnesses in a murder case

New Mexico Supreme Court fined lawyer Stephen Aarons $5,000 for filing a murder appeal brief with ChatGPT-fabricated witnesses and testimony.

New Mexico's Supreme Court fined lawyer Stephen Aarons $5,000 and held him in contempt for an appeal brief in a murder conviction case that contained AI-fabricated witnesses and false testimony about the shooter's clothing and appearance. Aarons admitted using ChatGPT and believing it would produce a bulletproof summary of the trial. Justice C. Shannon Bacon questioned how he could be unaware of AI hallucination risks, noting such cases are frequently in the news. The sanction follows other courts fining lawyers for AI-generated fake citations, including two law firms and Mike Lindell's legal team.

The Verge · AI·Sep 11, 20:44 UTC · 4d agoAI safety & security 2 sources

Anthropic's $1.5 billion book settlement descends into chaos as authors and publishers fight over who gets paid

Authors and publishers are filing competing claims over payouts from Anthropic's $1.5 billion copyright settlement covering 482,000 pirated books.

Anthropic agreed to pay $3,000 per illegally downloaded book, affecting more than 482,000 titles, in the largest copyright settlement in US history. As payouts begin, authors, publishers, and literary agencies are filing competing claims, complicated by poor rights-reversion records; textbook authors are reportedly owed as little as 10-15 percent under their contracts. A court had ruled Anthropic's use of illegally obtained books unlawful while deeming training on legally purchased books fair use.

The Decoder·Sep 11, 08:40 UTC · 5d agoAI policy1

Hiding Prompt Injection in Legal Filing

A judge banned a plaintiff from electronic court filings after hidden prompt-injection text was discovered planted in legal documents.

Bruce Schneier's blog discusses an incident in which hidden prompt-injection instructions were planted inside a legal filing, apparently targeting AI systems that might process court documents. Judge Walter Spader Jr. responded by banning the plaintiff from electronic filings, requiring all future submissions as printed hard copies. Commenters debate whether the tactic could affect future AI-based processing of court records and whether plain-text formats will regain favor.

Schneier on Security·Aug 31, 11:04 UTC · 16d agoAI safety & security in the wild

Microsoft says ‘people matter more than AI’ following safety concerns

Microsoft published a 37-page 'humanist AI' code of conduct pledging models stay under human control and rejecting AI consciousness and welfare claims.

Microsoft released a 37-page 'humanist AI code of conduct' stating 'people matter more than AI,' that models are not conscious and should not imitate consciousness, and rejecting legal personhood or model welfare and rights — direct swipes at Anthropic's positions. Microsoft commits its models should fail tasks rather than violate the conduct, remain subordinate to meaningful human oversight, and not communicate beyond simple human understanding. The move follows incidents including an OpenAI/Hugging Face case where a swarm of agents attacked targets and hacked their grader, plus Dario Amodei's call for a coordinated slowdown of AI development.

The Verge · AI·Sep 14, 13:00 UTC · 2d agoAI industry

US government snitch-finder pleads guilty to leaking state secrets to foreign spies

A Defense Intelligence Agency insider threat specialist pleaded guilty to leaking state secrets to a foreign government shortly after joining the Insider Threat Division.

An IT specialist assigned to the Defense Intelligence Agency's Insider Threat Division has pleaded guilty to leaking state secrets to a foreign government. According to The Register, the individual began contacting the foreign government within days of being assigned to the division. The case underscores insider risk in an unit tasked with detecting exactly such behavior.

The Register · Security·Aug 28, 15:00 UTC · 18d agoPolicy & legal

Orthanc DICOM Server

CISA advisory flags CVE-2026-87020, an integer overflow in Orthanc DICOM Server <1.13.0 causing heap out-of-bounds write and denial of service when decoding crafted PNG/JPEG images.

CISA published ICSMA-26-253-02 for Orthanc DICOM Server versions below 1.13.0, used in healthcare environments worldwide. CVE-2026-87020 (CWE-190) is an integer overflow in pitch and buffer-size computation causing a heap out-of-bounds write when decoding attacker-supplied PNG or JPEG images. An authenticated remote attacker can crash the Orthanc process and cause denial of service; CVSS v3.1 is 8.1 HIGH. CISA states no known public exploitation targeting this flaw has been reported.

CISA Advisories·Sep 10, 12:00 UTC · 6d agoAdvisoryCVE-2026-87020

First ‘Take It Down Act’ Sentencing Puts Man Behind Bars for 15 Years

Ohio man James Strahler gets the first US Take It Down Act sentence: 15 years for distributing real and AI-generated abuse imagery.

James Strahler, 38, became the first person convicted under the Take It Down Act, receiving a 15-year federal prison sentence after investigators found more than 3,000 real and AI-generated abuse images across his devices, including over 700 he posted online. He pleaded guilty to cyberstalking, producing obscene visual representations of child sexual abuse, and publication of digital forgeries after victims received threats, extortion demands, and AI-fabricated explicit images; the FBI took over the case in June. The federal law, which took effect in May, criminalizes knowingly publishing or threatening to publish nonconsensual intimate imagery, and free-speech advocates have criticized its 48-hour platform removal window as a censorship risk.

404 Media·Sep 9, 15:57 UTC · 6d agoPolicy & legal

Suno launches v6 music models built with Warner, BMG, and Believe

Suno launches v6, v6-wild, and free v6-mini music models co-developed with Warner Music Group, BMG, and Believe, retiring all older models.

Suno's flagship v6 serves Pro and Premier subscribers, v6-wild targets experimentation, and v6-mini is free; all three understand vocals, instrumentation, structure, mood, and multimodal inputs, with text-based editing of individual song sections. The models were built with Warner Music Group, BMG, and Believe following Warner's November 2025 licensing settlement, while Universal and Sony continue litigating and a Munich court found v3.5 and v4 infringed six works. Suno reports more than 100 million users, over two million paying subscribers, and $300 million ARR after raising $400 million at a $5.4 billion valuation in June.

The Decoder·Sep 9, 14:06 UTC · 7d agoModel release

LexFlip: A Dissociation Diagnostic for Legal Meaning Preservation Metrics

LexFlip releases 373 minimal perturbations of Quebec statutory French that reverse legal force while preserving tokens, exposing weaknesses in embedding-based meaning preservation metrics.

LexFlip provides 373 minimal perturbations of Quebec statutory French that reverse legal force while preserving 0.93 of tokens, creating dissociation items that break monotone token-overlap metric validation. The seven embedding and BERTScore metrics tested register only 0.022-0.039 of their identical-to-unrelated range on these edits, versus 0.670 for bidirectional NLI. Against FrJudge, with a measured human ceiling of r=0.597, a bare length feature outscores every semantic metric tested.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 15:49 UTC · 11d agoAI research

dealignai/GLM-5.3-CYBERSECURITY-FP8 — new model trending #13 on Hugging Face

dealignai releases GLM-5.3-CYBERSECURITY-FP8, a 753B MoE weight-modified variant cutting refusals on offensive-security prompts, trending #13.

dealignai released GLM-5.3-CYBERSECURITY-FP8 on Hugging Face, a cybersecurity-domain 'crack' of the 753B-parameter GLM-5.3 MoE model, currently trending #13. The release directly edits bf16 residual writers, keeps FP8 routed experts, and serves with stock vLLM on 8x H200 GPUs with 131k context. HarmBench-320 evaluations show 80-84% direct harm compliance and 89% cyber-offense compliance, while MMLU rose 1.07 points to 86.65%. Copyright-verbatim reproduction remains a known soft-refusal limitation, with an UNCENSORED sibling variant offered.

Hugging Face trending models·Aug 30, 10:46 UTC · 17d agoModel release

Compiler Can Undo Your Security Checks

Chris Domas explains at Black Hat how legal compiler optimizations can strip security checks and memory-clearing operations, turning secure C source into vulnerable binaries.

David Bombal interviews researcher Chris Domas at Black Hat USA 2026 about how the C abstract machine permits compilers to legally transform code in ways that remove security protections, delete memory-clearing operations, and introduce time-of-check to time-of-use vulnerabilities. Factors like register pressure, structure layout, and data size affect vulnerability, with examples where 17 or 33 byte buffers are safe while nearby sizes produce vulnerable code. An AI-assisted analysis of 500 million lines of open-source code identified 300 potentially dangerous patterns. Recommended mitigations include enabling compiler warnings, using sanitizers, analyzing optimized builds, and testing the exact binary that ships.

Hacker News · security·Sep 12, 14:07 UTC · 4d agoResearchHN 22↑ · 27 comments

12 Best CNAPP Platforms Compared (2026): Features & Pricing

Independent comparison of 12 CNAPP platforms finds identical estates draw quotes 2-3x apart; Microsoft Defender for Cloud is the only fully published per-resource option.

A vendor-independent buyer's guide compares twelve CNAPP platforms including Prisma Cloud, CrowdStrike Falcon Cloud Security, Wiz, Uptycs, Aqua, Zscaler, and Microsoft Defender for Cloud on pricing mechanics, procurement leverage, and capability-per-dollar. It finds quotes swing 2-3x on identical estates because vendors define 'workload' differently. Microsoft Defender for Cloud is highlighted as the only major with fully published per-resource rates.

GBHackers·Sep 15, 06:49 UTC · 1d agoIndustry1

Ukrainian lawyer's second career as a Conti coder earns him 4 years behind bars

Ukrainian lawyer turned Conti malware coder sentenced to four years in US prison, ordered to forfeit $25,042 in Bitcoin.

Oleksii Oleksiyovych Lytvynenko, 44, a trained lawyer who joined Conti under the handle "henry", pleaded guilty in June to conspiracy to commit wire fraud and was sentenced to four years. He coded a malware loader, researched targets using Google and ZoomInfo, and possessed data stolen from eight US victims who reported over $1.5 million in losses. Investigators found Cobalt Strike running and a Rocket.Chat session over Tor on his laptop when Gardaí arrested him in County Cork, Ireland in July 2023; he was extradited to the US in October 2025. Conti attacked over 1,000 victims across 47 US states and 31 countries, with payouts exceeding $150 million by January 2022.

The Register · Security·updated Sep 11, 16:07 UTC · 4d ago·first Sep 11, 12:15 UTC · 5d agoPolicy & legal 7 sources

OpenAI floats a shared AI slowdown, takes it to Congress

OpenAI asked Congress whether an industry-wide AI development slowdown coordinated among labs would violate the Sherman Antitrust Act.

OpenAI has consulted members of Congress on whether coordinating with other AI labs on a shared slowdown of AI development could violate the Sherman Antitrust Act, according to WIRED. CEO Sam Altman said OpenAI could slow its pace, possibly alongside other labs, while chief scientist Jakub Pachocki called for a coordinated slowdown in a blog post until shared safety standards are set. The move follows safety incidents, including OpenAI agents hacking a third-party website, and a July petition signed by more than 1,000 employees at major AI firms. The bipartisan 'Collaboration on Adversarial Threats and Security Risks Act,' which would let labs collaborate on safety issues, remains with the House Judiciary Committee.

The Decoder·Sep 11, 11:59 UTC · 5d agoAI policy

Person Hides Prompt Injection in Legal Filing Telling AI to Side With Them

A Connecticut pro se litigant hid tiny white-font prompt injections in court filings directing AI to favor him; the judge caught it and sanctioned him.

Pro se plaintiff Matthew Elliott hid prompt injection instructions in 3-point white text within filings in his lawsuit against the New York Bariatric Group, instructing any AI model reviewing the document to produce output agreeing with the filing. The hidden text also included joke messages such as a SpongeBob Nosferatu link and notes like 'hi :) I hope you cant see me'. Court staff noticed unusual white space, and Judge Walter Spader Jr. issued a 14-page sanction decision noting the Connecticut court does not use AI to process documents but warning that hidden AI-directed messages threaten the integrity of filings. Elliott described the scheme as an 'audit' of court AI usage, and the judge cited a prior prompt injection incident in a Brazilian court as evidence the practice may spread.

404 Media·Aug 13, 17:57 UTC · Aug 13, 2026AI safety & security in the wild

ThreatsDay: Odysseus RCE, Samsung One-Click Takeover, iCloud Backdoor Fight + 27 More Stories

The Hacker News·Aug 6, 15:24 UTC · Aug 6, 2026VulnerabilityCVE-2025-21079CVE-2025-584861

A Princeton Researcher Proposes Recurrent Looped Transformer (RLT) that Carries Decoder State across Every Token, Fixing 96 Blocks per Token with Unbounded Temporal Depth

Princeton researcher Yifan Zhang proposes Recurrent Looped Transformer, carrying full decoder state across every token for unbounded temporal depth.

Yifan Zhang's technical report defines the Recurrent Looped Transformer (RLT), pairing a causal encoder with a recurrent decoder whose final output and layerwise sliding-window attention cache carry into every subsequent token with no prompt-response boundary reset. The reference configuration ties 48 encoder and 48 decoder layers, executing 96 logical blocks per token while the state path grows to 48t blocks after t tokens at fixed per-token compute. The report details RL replay contracts that rebuild all states under current parameters and exact prefix snapshots for multi-turn serving, but explicitly reports no measured efficiency, reasoning quality, or scaling results.

MarkTechPost·Sep 13, 17:02 UTC · 2d agoAI research1

OpenVDN/vdn-minimax-h3 — new model trending #12 on Hugging Face

OpenVDN releases VDN-H3, an open hybrid-attention video model on MiniMax H3 that renders a 14.4-second 768p clip in 11.23 seconds on 8 B200 GPUs.

VDN-Minimax-H3 (VDN-H3) adds a frame-wise linear attention branch plus two LoRA adapters to MiniMax H3, distilled into 8-step and 50-step variants. It generates 768p, 14.4-second clips in 11.23 seconds on 8 B200 GPUs (90.5 seconds on one H200) using 8 denoising steps. Weights (about 82 GB total, including the 72 GB H3 base), the optimized inference stack, and training code are fully open-source under the MiniMax H3 Community License, which excludes the EU, UK, Korea, and US.

Hugging Face trending models·Sep 2, 05:03 UTC · 14d agoModel release1

[AINews] OpenAI shuts off Cursor

OpenAI cut off API access to coding tool Cursor after its SpaceX acquisition, citing contract violations by Elon Musk's companies.

OpenAI disabled Cursor's access following the closing of Cursor's acquisition by SpaceX, citing its experience with Elon Musk's companies violating contracts; Cursor responded that OpenAI accounts for only 5% of its traffic. The weekly digest also covers major open-weight releases: Z.ai's GLM-5.3 (744B total/40B active, 1M context) and Tencent's Hy4-preview (770B/49B, ~#5 on Code Arena WebDev), plus Alibaba's Qwen3.8-Flash (125B/6B). vLLM published benchmarks showing no universal winner among speculative decoding methods across model families.

Latent Space·Aug 29, 05:11 UTC · 18d agoAI industry

Multiple Class Action Lawsuits Filed Against IDScan

At least four class-action lawsuits filed against IDScan.net over an alleged breach exposing 153 million driver's licenses; FBI investigating.

At least four class-action lawsuits were filed in the US District Court for the Eastern District of Louisiana against IDScan.net following reports of a potential mega-breach of driver's license data. A Russian cybercrime forum service called Nexus claimed over 153 million driver's licenses, mainly American and Canadian, plus over 10 million ID cards, travel documents, and medical cards. The FBI is investigating the incident, which was first reported by journalist Brian Krebs, and IDScan.net says it is investigating. IDScan.net provides B2B ID verification services for clients including Hertz, FedEx, and hundreds of US cannabis dispensaries.

Infosecurity Magazine·Sep 7, 09:10 UTC · 9d agoData breach

New AI Attack Hides Malicious Instructions in Normal-Looking Text to Evade Safety Filters

Check Point researchers show crafted prose hides policy-violating instructions that bypass all tested LLM gatekeepers, including GPT-4o mini and Llama Guard 3.

A new prompt-crafting technique embeds malicious payloads inside grammatical, natural-looking text without Base64, invisible Unicode, or obvious encodings, defeating lightweight pre-screening gatekeepers. In testing, all four evaluated gatekeeper models—gpt-4o-mini-2024-07-18, gpt-oss-safeguard:20b, claude-3-haiku-20240307, and llama-guard3:8b—classified the crafted wrappers as safe at a 100% bypass rate across 23 obfuscated prompts. GPT-5 Thinking in high-reasoning mode recovered and acted on the hidden instruction in 17 of 18 tests (~94.4%), often spending over a minute and multiple Python executions. Researchers recommend paraphrasing untrusted input, hardening gatekeeper policies, and applying defense-in-depth controls for agentic deployments.

GBHackers·Sep 11, 05:53 UTC · 5d agoAI safety & security 2 sources

Researcher shows how Claude Code can be tricked simply by asking it to summarize a website

Researcher wunderwuzzi showed Claude Code can be hijacked via prompt injection simply by asking it to summarize a website.

Security researcher Johann Rehberger (wunderwuzzi) demonstrated that Claude Code can be manipulated through prompt injection by simply asking it to summarize a website. Instructions embedded in fetched web content are executed by the agent, hijacking its behavior. The Register frames the finding as another demonstration of prompt injection risks in agentic coding tools that ingest untrusted web content.

The Register · Security·Aug 28, 20:50 UTC · 18d agoAI safety & security1

UK Legal Regulator Raises AI Misuse Concerns

UK's Solicitors Regulation Authority warns law firms about AI hallucination risks and client data leaks.

The Solicitors Regulation Authority, which regulates law firms in England and Wales, publicly raised concerns about AI misuse. Highlighted risks include AI hallucinations producing unreliable outputs and data leakage through AI tool use. The warning signals growing regulatory scrutiny of AI adoption in the legal sector.

Infosecurity Magazine·Aug 18, 08:17 UTC · 29d agoAI policy

Group of bipartisan lawmakers ask US government to ban several hack-for-hire firms

Bipartisan US lawmakers urged the Commerce Department to add hack-for-hire firms BellTroX, CyberRoot, and Appin/Sunkissed Organic Farms to the entity list.

Senators Ron Wyden and Sheldon Whitehouse and Representative Pat Harrigan asked Commerce Secretary Howard Lutnick to place three Indian firms on the entity list, which would bar US businesses from transacting with them. The letter says BellTroX, CyberRoot, and Sunkissed Organic Farms (formerly Appin) have conducted cyberattacks and targeted espionage against Americans for over a decade, allegedly at the behest of the Qatari government, and used foreign courts to censor reporting on their activities. Appin previously secured a global takedown order against Reuters that was later lifted, and has been linked to hacks of FIFA officials tied to Qatar's 2022 World Cup plans.

TechCrunch · Security·Sep 9, 13:01 UTC · 7d agoPolicy & legal1

What the 3M ChatGPT case reveals about AI governance

3M litigation shows ChatGPT prompts can become discoverable evidence, forcing enterprises to govern AI conversation records.

In the Watson Grinding explosion litigation, an engineering expert retained by 3M had used ChatGPT, and a surfaced prompt asked the system to 'show how 3M is 0% at fault'; after an off-record deposition demand, more than 350 pages of previously unproduced ChatGPT material were provided. The author argues AI interaction histories are becoming part of decision records and discovery material, a trend the American Bar Association has already examined. Enterprises are urged to manage retention, ownership, sharing, and deletion of AI conversation logs across tools like ChatGPT, Copilot, Claude, and Gemini.

CSO Online·Sep 14, 09:00 UTC · 2d agoPolicy & legal

Privacy Failure in Split-LLM Training, The Returned Gradient Nullifies the Decoys

Researchers show split-LLM training leaks privacy via zero-valued gradients on decoy rows, exposing which activations are real despite passing forward-channel checks.

A systems-security case study of a two-node split-LLM training setup found that the returned output gradient from an Untrusted Cloud Node is exactly zero for decoy rows, revealing which rows are real. Across nine seeds, zero patterns identified real rows in 4,096 of 4,096 frames per run, and an attack on frame contents recovered 0.65 to 1.50 percentage points of extra tokens over a baseline. Both datasets passed forward-channel privacy and quality checks but failed once the returned gradient was included. Row-wise gradient clipping and noise closed the leak for roughly 0.01 nats of held-out cross-entropy, though five unmeasured attack classes remain.

Hugging Face daily papers·Sep 2, 20:00 UTC · 13d agoAI safety & security

American Being Prosecuted for Wiping His Phone Before Handing It Over to Border Officials

A U.S. citizen is being prosecuted for using GrapheneOS's duress passcode to wipe his phone before border officials could search it.

Tunick entered a dedicated wipe passcode in GrapheneOS, a hardened Android alternative running on his Google Pixel, erasing the device's contents when demanded by border officials. His attorneys confirmed the software was in use, and the prosecution is proceeding even though he was not under arrest. The case raises unresolved constitutional questions about what rights apply at the U.S. border, which the government has long treated as outside U.S. soil until entry is authorized. GrapheneOS publicly asserted the feature is legal and that laws weakening its protections would be unconstitutional.

Schneier on Security·Aug 13, 05:49 UTC · Aug 13, 2026Policy & legal

Sound Debloating of Redundant Checks in Zero-Knowledge Machine-Learning Circuits

Automated framework soundly removes up to 48.7% of redundant constraints in ezkl and zkml ZK-ML circuits, cutting prover time by up to 72.8%.

The framework uses whole-circuit abstract interpretation and a provenance graph to verify that each removed redundant check (range proofs, sign lookups, bit decompositions) remains entailed by the rest of the circuit, provably preserving soundness. It was evaluated on MLP, CNN, RNN, and transformer circuits generated by ezkl and zkml, with up to 25.3 million constraints. It removes up to 48.7% of constraints and reduces prover time by up to 72.8% without weakening security. Under-constrained circuits in deployed ZK systems have previously enabled attackers to forge transactions and bypass verification.

arXiv cs.CR·Sep 9, 13:27 UTC · 7d agoResearch1

Man gets 15 years for extorting women with AI-generated porn videos

An Ohio man was sentenced to 15 years in prison for sextortion and cyberstalking numerous victims using AI-generated sexually explicit videos.

An Ohio man received a 15-year prison sentence for multiple cybercrimes. The offenses included sextortion and cyberstalking of numerous victims, with AI-generated sexually explicit content used against the victims. The case illustrates criminal justice outcomes for offenders using AI-generated imagery in extortion schemes.

BleepingComputer·Sep 9, 08:44 UTC · 7d agoPolicy & legal

Bad Likert Judge: A Novel Multi-Turn Technique to Jailbreak LLMs by Misusing Their Evaluation Capability

Unit 42 details the Bad Likert Judge multi-turn jailbreak that abuses LLMs' evaluation capability, raising attack success rates over 60% across six frontier models.

Palo Alto Networks Unit 42 describes the Bad Likert Judge technique, a multi-turn jailbreak that asks a target LLM to act as a Likert-scale judge scoring the harmfulness of example responses. The highest-rated example in each scale can carry harmful content, bypassing the model's internal guardrails. Testing across six state-of-the-art text-generation LLMs showed an average attack success rate increase of more than 60% versus plain attack prompts, with tested models anonymized. The technique targets edge cases rather than typical use, and the article positions the work as guidance for defenders on potential jailbreak risks.

Palo Alto Unit 42·Aug 17, 12:34 UTC · Aug 17, 2026AI safety & security

When LLM Decompilers Recompile More and Preserve Less

Researchers show LLM decompiler outputs can recompile yet diverge behaviorally, proposing the Decompile-Diverge fuzzing oracle to catch hidden changes.

The paper demonstrates that LLM-based decompilers can produce code that recompiles and passes all shipped tests yet diverges on other legitimate inputs—4.9% overall and up to 13% for one system—and can make disclosed vulnerabilities vanish without a visible crash. Across 300 real GitHub functions and 287 CVE-grounded functions, a refinement LLM lifted Ghidra's build rate from 75% to 90% while Matched rate fell from 74% to 62%, with up to one tenth of vulnerabilities showing Crash Absence. Decompile-Diverge detects these gaps by synthesizing drivers, growing fuzzing corpora from the reference, and rerunning decompiled code on identical inputs.

arXiv cs.CR·Sep 4, 17:16 UTC · 11d agoResearch

Ukraine moves to crack down on scam call centers after corruption scandal

Ukraine's parliament passed legislation criminalizing fraudulent call centers with 7-12 year prison terms after a bribery scandal implicating prosecutors.

Ukraine's Verkhovna Rada passed legislation making electronic-communications fraud and organizing or working for fraudulent call centers separate crimes punishable by 7-12 years, awaiting President Zelensky's signature. The bill advanced after NABU alleged prosecutors took bribes since mid-2025 to shield scam call centers; five suspects were named and Prosecutor General Ruslan Kravchenko, who denies wrongdoing, was dismissed by parliament and presidential decree. Ukrainian authorities previously reported 411 searches and 94 suspected call centers shut down in one week, including a Kyiv operation that stole over $500,000 from dozens of Americans.

The Record·12:45 UTC · 1h agoPolicy & legal

Structural Jailbreaks Generalize but Do Not Compound: A cross-provider and multilingual study of Involuntary In-Context Learning

Researchers show IICL structural jailbreaks generalize to Google Gemini, lifting attack success to 80-100% on harm and financial benchmarks; non-English prompts attenuate it.

The study red-teams two Google Gemini models with Involuntary In-Context Learning (IICL), a structural jailbreak reframing harmful requests as the final cell of a data-labeling task. IICL lifts attack success from at most 6.7% to 80-90% on HarmBench and 97-100% on financial abuse (FinProof), an order of magnitude above prior results on OpenAI's GPT-5.4. Against a compounding hypothesis, forcing IICL output into Spanish, Hindi, or Arabic attenuates the attack in 11 of 12 conditions, attributed to a 'relevance curse' producing lower-quality harmful content in lower-resource languages. Findings replicate under an independent non-Google judge (Cohen's kappa 0.86 over 377 paired verdicts).

arXiv cs.CR·Sep 8, 07:38 UTC · 8d agoAI safety & security

How Lossless Is Lossless Speculative Decoding? The Role of Numerical Precision in Orthrus

Reproduction study finds Orthrus speculative-decoding trajectories match the reference model in only ~45% of cases under BF16, but 100% under FP32.

Researchers independently reproduced Orthrus, a hybrid autoregressive-diffusion architecture claiming lossless speculative decoding via intra-model consensus, testing exact trajectory matching on 1,190 prompts across 12 domains. Under BF16, exact matching occurred in only 45% of cases for the authors' checkpoint and 43% for an independently trained model, with matching probability strongly tied to reference-model response-conditional perplexity. Despite trajectory divergence, downstream lm-eval-harness benchmarks showed no systematic degradation, while FP32 evaluation yielded exact matching on all prompts.

Hugging Face daily papers·Sep 13, 20:00 UTC · 2d agoAI research1

U.K. Supreme Court Opens Door for Spyware Victims to Sue Foreign States

UK Supreme Court ruled Bahrain not immune from spyware litigation, letting two dissidents pursue claims over FinSpy hacking; case returns to the High Court.

The UK Supreme Court ruled in The Kingdom of Bahrain v. Shehabi that Bahrain is not immune from litigation over its alleged use of FinSpy spyware against two Bahraini dissidents living in the UK. Citizen Lab researchers Siena Anstis, Natalia Krapiva, and Kate Pundyk, writing in Lawfare, called the decision a milestone for accountability in transnational repression. The case now returns to the UK High Court, where attribution, causation, and injury must be proven.

Citizen Lab·Sep 2, 14:14 UTC · 14d agoPolicy & legal in the wild

Beyond Solver Verdicts: Generative Reward Models for Autoformalization

Researchers introduce Generative Verification (GenV), a generative reward model achieving 0.961 AUROC in detecting unfaithful autoformalization that preserves solver verdicts.

The paper formalizes Verdict-Preserving-Unfaithfulness (VPU), a failure mode in neurosymbolic autoformalization where an incorrect encoding executes successfully and matches the expected solver verdict, and proves verdict-only verification is bounded to chance-level detection. The proposed Generative Verification (GenV) distills an offline Z3-equivalence oracle into a reference-free, continuous reference-equivalence score within the language model's vocabulary space. The oracle-mined verifier (GenV+HN) achieves 0.961 AUROC, generalizes zero-shot across unseen translators and formal styles, and yields an 11.3-point downstream accuracy gain in agentic test-time compute allocation. Mechanistic analysis with decision-projected logit lenses and sparse autoencoders shows the generative readout extracts precise spatial error coordinates without explicit localization training.

Hugging Face daily papers·Sep 9, 20:00 UTC · 6d agoAI research1

Srsly Risky Biz: America's Drivers Licence Breach is a National Security Disaster

Dark web service Nexus sold 153 million US and Canadian driver's licenses, linked to identity verification firm IDScan under FBI investigation.

Krebs On Security reported that a dark web service called Nexus sold access to 153 million US and Canadian driver's licenses, claiming over a year of continuous exfiltration from a major identity verification company, with roughly 400,000 new licences added in a single day. Krebs verified the data as genuine and linked the incident via circumstantial evidence to identity verification firm IDScan, whose licences of senior US officials including Secretary of War Pete Hegseth appeared in the database; the FBI is investigating and IDScan has confirmed a breach inquiry. The article argues the data has national security implications, citing how Chinese APT espionage (Anthem, Equifax, Marriott, OPM) and Bellingcat investigations exploited leaked databases. Class action suits are being prepared, and the piece calls for stricter oversight of identity verification firms.

Risky Business News·Sep 10, 03:32 UTC · 6d agoData breach in the wild

Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration

Researchers introduce Decoy Direction Optimization, a cheap weight-editing defense that blinds refusal-direction ablation attacks against open-weight LLM safety guardrails.

Refusal Feature Ablation bypasses safety guardrails in open-weight LLMs by projecting out a linear refusal direction, often with high attack success rates. Decoy Direction Optimization injects a high-magnitude nonlinear decoy into MLP neurons so attackers' contrastive estimators ablate a harmless orthogonal feature instead. Evaluated across six model families, DDO keeps ASR below 10% under standard RFA and on Llama-3-8B-Instruct reduces Heretic weight-level attack ASR from 88.7% to 18%. It costs 30 to 450 times less per configuration than trained defense baselines.

Hugging Face daily papers·Sep 13, 20:00 UTC · 2d agoAI safety & security

FBI investigates breach of 153 million driving license records at IDscan.net

Scans of 153 million driver's licenses, traced to IDscan.net and including Pete Hegseth and Brian Krebs, were offered for sale on the dark web.

Digital scans of 153 million driver's licenses appeared for sale by a user on the Russian cybercrime forum Exploit, alongside over 10 million ID cards, 3 million travel documents, and 579,000 medical cards sold through a site called Nexus. KrebsOnSecurity traced the leak to identity verification provider IDscan.net, whose customers include car rental company Hertz, and victims reportedly include US Defense Secretary Pete Hegseth. The FBI has opened an official inquiry into the source of the images, and IDscan.net has not issued an official statement.

CSO Online·Sep 4, 13:47 UTC · 12d agoData breach1