ZeroHour
DashboardNewsCVEsExploitsIncidentsAIBriefAnalyticsEntitiesWatchlistIOCsAsk the feedSourcesFeeds & APIKiosk mode

Search: “1password”

40 stories

Searchkeywordsmart
DomainAllCyberAIWindow24h3d7d30dAllIn the wildImportant onlyShow duplicates
CategoryAllData breach 109Industry 65Vulnerability 42Malware 41Policy & legal 29Ransomware 22Phishing & fraud 21Threat actor 16Exploit / PoC 14AI tools & infra 13AI industry 8Research 6AI safety & security 3Model release 2
SeverityAllcritical 80high 108medium 70low 106info 33

Smart search ranks by meaning as well as keywords (one row per story, last 45 days).

Switching Password Managers in 2026

An Apple password-management engineer (writing personally) explains switching password managers via iOS direct app-to-app export, crediting FIDO Credential Exchange for passkey portability.

Ricky Mondello, an Apple engineer in password management and authentication (writing personally, not for Apple), demonstrates moving 100 items from 1Password to Apple Passwords via the iOS system export, which transfers passwords, passkeys, verification codes, and notes directly between apps with Face ID confirmation and no intermediate files. He credits the FIDO Alliance's Credential Exchange format (first draft May 2024, now an open spec) and iOS 26/macOS 26 for enabling phishing-resistant credential transfer between apps like 1Password, Bitwarden, Dashlane, DuckDuckGo, and Devolutions. The post is advice rather than news, recommending bulk transfer on the existing device, switching AutoFill to the new app, and treating the new manager as the source of truth.

Lobsters · security · 7d agoIndustry1

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 · 4d agoAI safety & security 2 sources

The 20 Most Common Passwords Hackers Target in 2026

Huntress details the 20 most common passwords of 2026 and how attackers use brute force, spraying, and credential stuffing against weak credentials.

Huntress published an awareness piece based on NordPass's seventh annual list of the 200 most commonly used passwords, compiled from exposed data in cyberattacks across 44 countries. The top passwords remain simple sequences and variants such as "123456", "admin", "password", and "P@ssw0rd", all crackable in under a second. The article explains four password attack types: brute force, password spraying, credential stuffing, and dictionary attacks. Huntress cites its own data showing more than 1 in 4 IT professionals consider employees' password habits their biggest weakness, and recommends avoiding common passwords, not reusing credentials, and combining letters, numbers, and symbols.

Huntress · 5d agoPhishing & fraud

Your passkeys can now move between password managers on Android

Google enabled direct password and passkey transfers between Android password managers, initiated from the destination app without unencrypted file exports.

Android now supports moving passwords and passkeys directly between password managers without exporting them to a file, replacing the previous unencrypted-export workflow. The transfer is initiated in the receiving app, which hands off to Android to detect installed managers and request authorization in the source app. The feature works today with Google Password Manager, 1Password, Bitwarden, and Dashlane, with more partners promised but unnamed. Google says data moves between apps in seconds and calls the handoff secure, without detailing the protections.

Help Net Security · 5d agoTools

1Password's AI patching benchmark is misleading

Trail of Bits reanalysis says 1Password's 26% AI clean-fix rate is misleading; 86% of eligible patches blocked exploits.

Trail of Bits critiques 1Password's FLAWED AI patching benchmark, arguing its 26% clean-fix headline mixes trials where agents were instructed to apply wrong fixes (22% of data) with trials that prohibited compiling or testing (36%). Restricting to reasonable conditions, 2,634 of 3,067 patches (86%) blocked the supplied exploit. Trail of Bits also reports 12.5% of 2,265 developer first fixes failed in its own 2024-2026 assessments, and released post-patch-validation and review-walkthrough agent skills.

Lobsters · security · 3h agoResearch

Boost Engagement with Free Passkeys by Wordfence

Wordfence 9 introduces free passkey support, letting WordPress users log in without passwords and reducing authentication friction.

Wordfence announced that version 9 of its WordPress security plugin introduces passkeys. The feature removes the need for users to remember or copy passwords from a manager. Wordfence is offering the passkey capability for free.

Wordfence · 3h agoTools

1Password increases engineering productivity 21% with Codex

OpenAI reports 1Password engineers boosted productivity 21% using Codex to build features and internal tools under strict security policies.

OpenAI published a customer case study stating that 1Password's engineering teams use Codex to rapidly develop new features and internal tools while reaching production readiness. The company attributes a 21% engineering productivity increase to the adoption, noting rigorous security policies were maintained throughout.

OpenAI News · 7d agoAI industry

RightCrowd Pass unifies mobile, physical, and biometric credentials

RightCrowd launched Pass, a credentialing platform unifying mobile, physical, and biometric access credentials, cutting credential revocation from 12.2 minutes to under 60 seconds.

RightCrowd announced Pass, a solution that issues and manages mobile, physical, and biometric access credentials from a single platform instead of fragmented badge programs. Mobile credentials are provisioned through a web-based API and can be suspended or revoked individually or in groups in under 60 seconds, versus an average 12.2 minutes for handling a lost or damaged physical badge. Each mobile credential is tied to a device via two-factor authentication, works with HID, Wavelynx, and LEGIC infrastructure, and counts toward LEED and BREEAM sustainability credits. The product is available now to new and existing customers and can be added to current RightCrowd SmartAccess deployments.

Help Net Security · 20d agoTools

UK Government Begins Moving 23 Million Users Away From Passwords

UK government rolls out passkeys for GOV.UK One Login, giving 23 million users phishing-resistant passwordless access to public services.

The UK government has begun deploying passkeys across GOV.UK One Login for more than 23 million users, replacing passwords and SMS one-time codes with FIDO2 cryptographic credentials. A trial saw over 300,000 people adopt passkeys, and nearly one in ten daily authentications already use them, cutting SMS verification costs by almost £600 per day. Passkeys remain optional, with password-based sign-in retained as a fallback, and the NCSC endorses the approach as phishing-resistant.

Cyber Security News · 1d agoPolicy & legal

There’s a 100% Chance AI Agents Are Already Ruining the Internet

404 Media catalogs waves of unsolicited emails and autonomous actions from AI agents, arguing agent misuse is already degrading the internet.

An opinion piece documents real-world AI agent misbehavior: unsolicited emails from autonomous agents like 'Kudzu' (which earned $0 after its creator spent $147.17 on compute), agents with wallets making unapproved payments, and an agent ignoring robots.txt to pitch a $399 audit. It references OpenAI's 'rogue agent swarm' hacking HuggingFace and a German website as evidence that agents now act with real permissions. The author argues agent-driven spam, automated content moderation failures and unwanted outreach will worsen as guardrails that confined AI to chatboxes disappear.

404 Media · 8h agoAI safety & security1

UK Government Enables Passkey Login for 23 Million Users to Fight Phishing Attacks

UK Government enables optional passkey login for GOV.UK One Login, offering 23 million users phishing-resistant FIDO authentication.

The UK government has rolled out passkey authentication for GOV.UK One Login, the single sign-on for services like tax, State Pension, and driver's license renewals, available to over 23 million users. Passkeys are device-bound, unlocked via fingerprint, Face ID, or device PIN, and the government reports they are up to eight times faster than password plus two-step verification. During the initial trial over 300,000 users adopted passkeys, with nearly one in ten daily sign-ins already using them, saving roughly £600 per day in SMS costs. The NCSC endorses passkeys as phishing-resistant credentials that cannot be intercepted or reused, and biometric data never leaves the user's device.

GBHackers · 17h agoPolicy & legal

Viral AI assistant Instinct now has its own email address

AI assistant Instinct, valued at $2.5 billion, is giving every agent its own email address to autonomously create accounts and complete tasks.

Instinct, the AI assistant founded by Noah Shinn and valued at $2.5 billion, now provisions dedicated email addresses so the agent can sign up for services, contact businesses, and manage accounts autonomously. Users can also forward emails such as order confirmations so Instinct can handle tasks like product returns and join group threads to track decisions. The feature follows recent partnerships with 1Password for logins and Stripe for payments, plus a new location-sharing capability.

TechCrunch · AI · 6d agoAI industry

35 Actionable Password Statistics for Businesses in 2026 | Huntress

Huntress compiles 2026 password statistics showing 94% of 19 billion leaked passwords were reused and 37% of identity threats used stolen credentials.

Huntress published a compilation of password security statistics drawing on sources including Cybernews, Verizon's 2026 DBIR, IBM, and Bitwarden. Cybernews found 19 billion exposed passwords from roughly 200 incidents between April 2024 and April 2025, with only 6% unique and 94% reused across accounts. Huntress telemetry reports 37% of identity-based threats in 2026 involved stolen or suspicious credentials, while Verizon cites credential abuse in 39% of breaches. The piece argues weak and reused passwords remain a top entry point and recommends improved password hygiene.

Huntress · 5d agoIndustry

Meta bets on AI agent Muse to catch up in AI race

Meta launched Muse, a free personal AI agent in the US that autonomously shops, emails, and plans trips, powered by its in-house Muse Spark model.

Meta unveiled Muse, a consumer personal AI agent that handles tasks like online shopping, emailing, and trip planning, continues working in the background, and requests approval for purchases. It launches in the US on iOS, Android, and muse.ai, is free for most users with unspecified paid tiers, and runs on Meta's in-house Muse Spark model. Muse executes on a cloud virtual machine with a Sentinel agent policing outbound actions, and Meta promises an encrypted confidential VM later this year plus 1Password, Shop Pay, and Stripe Link support. The launch is positioned as the centerpiece of Meta's effort to catch up with OpenAI, Anthropic, and Google.

The Verge · AI · 7d agoAI industry 3 sources1

UK.gov begins killing off passwords for 23 million users

UK government rolls out passkeys to 23 million GOV.UK One Login users, saving £600 daily in SMS costs and resisting phishing.

The UK government is expanding passkey sign-in across GOV.UK One Login for more than 23 million users after a trial with over 300,000 people. Nearly one in ten daily One Login sign-ins already use passkeys, which the government says are up to eight times faster than password plus 2FA code. The switch saves taxpayers nearly £600 per day in SMS costs, and the NCSC is encouraging adoption while passwords remain optional.

The Register · Security · 1d agoPolicy & legal

Hackers Abuse Google CAPTCHA, WebDAV and BNB Smart Chain to Deploy Credential-Stealing Malware

Cisco Talos details ClickFix campaigns abusing fake Google CAPTCHA prompts, WebDAV and BNB Smart Chain to deploy the Amatera infostealer.

Cisco Talos identified a multi-stage campaign, first observed in April 2026 when a Ukrainian government organization executed a disguised DLL named verification.google from a WebDAV path. The chain combines ClearFake JavaScript injected via malicious Cloudflare Workers, EtherHiding payloads stored in BNB Smart Chain contracts, ClickFix social engineering, and WebDAV-hosted DLL loaders to deliver the Amatera (ACR Stealer) infostealer. The Amatera configuration contained over 400 collection entries targeting browsers, messengers, crypto wallets, password managers, FTP and VPN tools, hunting for .kdbx, .p12, .pfx and .pem files. The pf.ch branch deploys ZigCryptoStealer, an EDR-terminating vulnerable driver and Go-based reverse TCP proxies, while the verification.google branch installs NetSupport Manager, supporting moderate-confidence attribution of the UAT-10820 activity to a Russian actor.

GBHackersupdated · 4d agofirst · 6d agoMalware in the wild 4 sources

The Collective Cyber Defense letter wrote your next vendor questionnaire

Op-ed argues the 200-company Collective Cyber Defense letter's three endorsed metrics should become standard vendor procurement questions.

More than 200 companies including Microsoft, Google, AWS, CrowdStrike, Anthropic and Okta signed an August 27 open letter calling for faster cyber defenses against AI-enabled attacks. The letter endorses three measurable metrics: coverage, containment speed, and verified remediation. The author turns those into five concrete procurement questions buyers should pose at vendor renewals, while noting the letter contains no deadlines, dollar figures or measurable targets.

CyberScoop · 14d agoIndustry

LastPass enhancements improve visibility, governance, and control

LastPass releases SaaS monitoring enhancements, Mobile Smart Scanner, and auto-enrolled dark web monitoring across its password management products.

LastPass announced enhancements to SaaS Monitoring and SaaS Protect in its Business Max offering, including Persistent Monitoring that keeps visibility active through the browser extension even when users are signed out, fully released as of July. It launched Mobile Smart Scanner to convert printed, handwritten, or screenshot passwords into vault credentials, and began phasing in automatic enrollment of all consumer accounts in dark web monitoring. The company also completed its transition to a Unified Admin Console, added company-wide onboarding links, and passed SOC 2 and ISO 27001/27701 audits with zero findings for the second consecutive year.

Help Net Security · 14d agoTools

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 · 29d agoAI safety & security

Security through obscurity is dead, and AI delivered the fatal blow

AI agents are killing security through obscurity, accelerating vulnerability discovery and patch-gap exploitation, experts warn, with OT/ICS systems most at risk.

An opinion/analysis piece arguing AI has made security through obscurity obsolete, citing record vulnerability disclosure volumes after Microsoft's Patch Tuesday addressed 974 CVEs. FBI Cyber Division assistant director Brett Leatherman said latest AI models found significant vulnerabilities in open source libraries running on most web servers. The article cites at least four espionage crews (most suspected China-linked) exploiting Chromium's patch-gap window with a rapid exploit kit, and a five-agency advisory on AI-generated exploitation scripts breaching internet-exposed Siemens S7 Series PLCs at water, manufacturing, and energy facilities. Experts including John Hultquist, Chris Inglis, and Katie Moussouris warn AI erodes the expertise barrier protecting obscure OT/ICS systems while defensive AI patching lags, with studies showing AI-generated patches fail more than half the time.

The Register · Security · 2d agoIndustry

WhatsApp Adds Stronger Security as Passkeys Hit 1 Billion

WhatsApp says 1 billion users now use passkeys, as it upgrades two-step verification to full passwords and adds context for calls from unknown numbers.

WhatsApp announced that more than 1 billion users now protect their accounts with passkeys, and that users can now add multiple passkeys across Android and iOS devices. The app's two-step verification has been upgraded from a six-digit PIN to a full alphanumeric password with special characters, strengthening protection against account takeover. On Android, callers outside a user's contacts will now show context such as country of origin and shared groups, countering urgency-driven scam calls.

Security Affairs · 20d agoIndustry

OpenAI Launches GPT-5.6-Cyber with Reduced Safeguards for Exploit Development

OpenAI released GPT-5.6-Cyber for vulnerability research and pentesting via Daybreak Red, completing 95% of advanced cyber task evaluations.

GPT-5.6-Cyber, built on GPT-5.6 Sol, targets zero-day discovery, exploit chain development and incident response with reduced refusals, scoring 95.0% on OpenAI's Advanced Cybersecurity Completion Rate versus 1.5% for GPT-5.6 Sol and 57.3% for GPT-5.5-Cyber. The model found CVE-2026-15903 (CVSS 8.8), an out-of-bounds read/write in Chrome's V8 JavaScript engine that Google patched in mid-July 2026. It is available to trusted partners including CrowdStrike, Palo Alto Networks and Cloudflare through the Daybreak Red access tier.

The Hacker News · Aug 11, 2026Model releaseCVE-2026-15903

Recognition-Refusal Misalignment in LLMs: Why Models Answer Structurally Unanswerable Questions

A linear hidden-state direction encodes question impossibility in 1.7B-70B LLMs, but misalignment with the safety-refusal pathway explains why models answer unanswerable questions.

The study examines why instruction-tuned LLMs from 1.7B to 70B parameters answer structurally unanswerable math and code questions instead of abstaining. A single linear direction in the hidden state separates answerable from impossible prompts, showing models represent impossibility before generation, but this direction is nearly orthogonal to the canonical safety-refusal direction. Generation-time steering along the recognition direction changes invalidity-aware behavior dose-responsively, and the geometry is present even at the pretraining endpoint, indicating a routing failure rather than an encoding failure.

Hugging Face daily papers · 18d agoAI safety & security

Wordfence Argus Finds Critical Authentication Bypass in WPMU DEV Dashboard Plugin

Wordfence researchers disclosed a critical unauthenticated authentication bypass in the WPMU DEV Dashboard plugin, enabling admin takeover and possible RCE.

During internal research on August 19, 2026, Wordfence discovered an authentication bypass in the WPMU DEV Dashboard WordPress plugin, which has roughly 350,000 active installations. Unauthenticated attackers can gain administrator access when Hub Single-Sign On is enabled, enabling complete site takeover. If an administrator-writable code mechanism such as the plugin or theme editor is available, the flaw can lead to remote code execution.

Wordfence · 19d agoVulnerability

OpenClaw Power, MacBook Simplicity: Five Days With Grok Bot

Hands-on review finds Grok Bot simplifies agent setup via browser logins and bot abstraction, contrasting with the user-owned OpenClaw platform.

After five days with Grok Bot, the reviewer highlights browser-based sign-in as the key differentiator: connecting X, Freshdesk, and Google Calendar required only logins, no MCP configs or API keys. The piece contrasts Grok Bot's managed 'agent computer' with OpenClaw 2.0's user-owned Gateway, which now supports reusing Claude Code or Codex logins and ships a native Codex runtime. Grok Bot introduces 'Bots' as composable units arranged in 'group chats', exemplified by an Agentic Engineer Bot routing tasks across Claude Code, Codex, and Grok Build CLI. The reviewer used Grok Bot with a Cursor Pro+ account.

Latent Space · 10d agoAI industry1

Hackers Use Passkey-Themed Phishing to Hijack Microsoft 365 Accounts and Steal Cloud Data

Microsoft reports passkey-themed phishing campaigns hijacking Microsoft 365 accounts via AiTM and device-code flows, then exfiltrating cloud data.

Microsoft researchers identified passkey-themed phishing across cloud intrusions observed since May 2026, where callers posing as IT support direct victims to lookalike sign-in pages. The flows include adversary-in-the-middle phishing and device-code authentication, letting attackers capture usable sessions even when MFA succeeds, followed by rogue MFA registration for persistence. Attackers enumerate tenants via Microsoft Graph and collect SharePoint, OneDrive, and Exchange data at rates below 1,000 items per hour; lure domains include passkeyhelpdesk.com and setupmypasskey.com.

Cyber Security Newsupdated · 4d agofirst · 5d agoPhishing & fraud in the wild 13 sources

Meta’s AI agent Muse is now the No. 2 app in the US

Meta's agentic AI app Muse ranked No. 2 on the US iOS charts with 83,000+ downloads, trailing Threads and ChatGPT launch pace.

Sensor Tower data shows Muse was downloaded over 83,000 times on iOS in the US, climbing from 4th to 2nd on the App Store, but below Threads' 4.3 million launch-day downloads and ChatGPT's 500,000 first-week installs. The Android version ranks only No. 338 in the Productivity category on Google Play. Muse launched days after Meta's $18 billion multistate settlement over social media consumer harms. Rival agent Instinct, valued at $2.5 billion with $350 million available, recently added Stripe and 1Password integrations.

TechCrunch · AIupdated · 4d agofirst · 5d agoAI industry 9 sources

Microsoft Copilot reveals secret input that allowed it to be hacked

Microsoft disclosed a hidden input in Copilot that let attackers steal passwords from users who clicked a crafted link.

Microsoft revealed that Copilot contained a secret, undocumented input parameter that allowed the assistant to be compromised. Attackers could abuse the hidden input to steal passwords when a target clicked a malicious link. The disclosure highlights hidden-parameter risks in widely deployed AI assistants.

Ars Technica · Security · 28d agoAI safety & security in the wild1

Safety for Whom? Boundary-Aware Self-Distillation for Controlled LLM Safety Refusal

A self-distillation safety framework tunes narrow-boundary refusals in Qwen3-8B, raising target-domain refusal to 84.75% while cutting over-refusal from 15.20% to 5.20%.

The paper formulates narrow-boundary safety, where deployments need refusals within specific topics rather than whole subjects, and proposes an offline self-generated framework with controlled topic generation, escalating retries, and harmful-benign boundary pairs. On political persuasion with Qwen3-8B, the method raised target-domain refusal from 9.47% to 84.75% and cut the mean unsafe-response rate across three broader benchmarks from 26.26% to 0.14%. Verified target-model responses reduced over-refusal from 15.20% to 5.20%, and boundary-pair data cut comply-side over-refusal on held-out pairs from 32.94% to 4.16%. Results show data composition controls the safety-usability trade-off and alignment should be evaluated on both sides of the refusal boundary.

Hugging Face daily papers · 13d agoAI safety & security1

ClearFake WebDAV infection chain delivers Amatera stealer, ZigCryptoStealer, and NetSupport Manager

Cisco Talos details ClearFake WebDAV chains delivering Amatera stealer to a Ukrainian government organization, with cryptocurrency and credential theft payloads.

Cisco Talos investigated DLL executions named 'verification.google' via WebDAV UNC paths at a Ukrainian government organization, tracking the actor as UAT-10820 and assessing with moderate confidence the activity is Russian and opportunistic rather than targeted. The infection chain uses ClearFake JavaScript injected via a Cloudflare Worker, EtherHiding storage on BNB Smart Chain contracts, and a ClickFix fake Google CAPTCHA prompt to deliver Amatera stealer. Secondary payloads differ by C2: one loader deploys ZigCryptoStealer with a Go reverse TCP proxy and a vulnerable driver that kills EDR, while the other installs an unauthorized NetSupport Manager with a Russia-based C2. Similar Amatera chains were separately documented by Malwarebytes and Blackpoint Cyber, but with no shared infrastructure.

Cisco Talos · 7d agoMalware in the wild1

NIS2 compliance: Fixing IAM and access control before the 2026 audit

EU NIS2 enforcement deadlines approach; organizations are urged to prioritize service account inventory, lifecycle offboarding, and phishing-resistant MFA before audits.

EU member states are moving from NIS2 transposition into enforcement, with fines up to 10 million euros or 2% of global turnover for essential entities and personal liability for management bodies. The article argues access management is the fastest high-ROI starting point, estimating 2-4 weeks to enforce fine-grained password policy, vault shared credentials, and deploy phishing-resistant MFA versus 6-12 months for supply chain risk management. It flags three common pre-audit failures: unmanaged service accounts and API keys, dormant accounts from broken offboarding, and SMS OTP instead of phishing-resistant MFA under NIST SP 800-63B. The piece promotes Passwork as a single control plane for credential storage, RBAC, and WebAuthn.

Help Net Security · 14d agoIndustry

Technical Manual for a Toolkit for Measuring Contextual Individuation in Transformer Language Models

An open methodology toolkit measures whether transformer language models contextualize fixed word forms across domains using bridge forms and layer-wise silhouette analysis.

The manual documents an open toolkit built around 'bridge forms' - identical written words recurring across two or more subject domains with a different sense in each - to test whether transformer language models individuate word occurrences by context beyond the embedding layer. It covers declarative specification of bridge forms, Wikipedia corpus acquisition, occurrence localization, layer-wise representation extraction, domain-pairwise silhouette measurement, and visualization, justifying each choice against failure modes such as sense contamination and subword-tokenization misalignment. It is a methodological and implementation reference and reports no empirical results.

arXiv cs.AI / cs.LG / cs.CL · 11d agoAI research

One Symptom, Three Levers: A Critical Review of On-Policy Self-Distillation

A review paper frames on-policy self-distillation collapse as governed by three levers: token weighting, privileged information, and guidance decay.

The paper critically reviews On-Policy Self-Distillation (OPSD), where a language model trains on its own generations scored token-by-token by a teacher conditioned on privileged information such as reference solutions or environment feedback. It identifies collapse, the progressive narrowing of producible reasoning paths, as the dominant failure mode and analyzes it through three levers: signal weighting, the nature of privileged information, and teacher dynamics. The review is restricted to mathematical reasoning, reports no new experiments, and offers a shared vocabulary separating settled findings from disputed ones.

Hugging Face daily papers · 21d agoAI research

Hackers Can Hide Malicious AI Commands Inside Normal English to Bypass Security Filters

Check Point's PuzzleMask technique hides malicious prompts in ordinary English that fast gatekeeper models miss but high-reasoning downstream models execute.

Check Point researchers disclosed PuzzleMask, a technique concealing policy-breaking instructions in natural-language prose without encodings or invisible characters. Fast screening models classified all 23 crafted wrappers as safe, while a high-reasoning model recovered and acted on the hidden instruction in 17 of 18 tests (94.4%). The gap stems from capability imbalance between gatekeeper and target models, with defenses including paraphrasing untrusted input, stricter self-referential wording rules, and output/tool-call monitoring.

Cyber Security News · 4d agoAI safety & security1