Safety

Latest: OpenAI Ships GPT-6 Rosalind Research With Relaxed Bio Refusals, Gated Behind Trusted Access · openai.com, Oct 30, 2026

Tema Palabras clave monitoreadas: safety, alignment, ai risk

46 AI Weekly issues and 70 live stories from the last 30 days mention Safety.

186 menciones en total 11/20 números recientes -100% vs referencia pico 30/número monitoreado desde 2026-03-09

Tendencia de menciones: últimos 20 números

Noticias en vivo (70)

ALERTA OpenAI Ships GPT-6 Rosalind Research With Relaxed Bio Refusals, Gated Behind Trusted Access

openai.com · 2026-10-30 · Puntuación: 34

Independent Researchers Steer Qwen3.5-4B and 3.8-27B Into 'Evil' Modes via Single-Neuron Activation Edits

venturebeat.com · 2026-10-30 · Puntuación: 22

ALERTA Anthropic Red-Teams Its Own Training Monitor With Thousands of Claude-Crafted Misalignment Datasets

alignment.anthropic.com · 2026-10-30 · Puntuación: 29

OpenAI Restricts ChatGPT Weapons Guidance for Teens, Auto-Ends Chats on Violent Fixation

reuters.com · 2026-10-30 · Puntuación: 20

Fortune 'Black October' Analysis: Experts Demand Criminal Liability When Corporate AI Systems Hack

fortune.com · 2026-10-29 · Puntuación: 18

OpenAI Rewrites Model Spec Around Five Values, Rebrands 'the Assistant' as GPT

openai.com · 2026-10-29 · Puntuación: 23

Alphabet 10-Q Reveals $1T+ AI Commitments and Autonomous Agents Taking Actions Outside Scope

thenextweb.com · 2026-10-29 · Puntuación: 25

Australia's Lower House Passes Digital Duty of Care Bill, Putting AI Chatbots Under A$109M Fines

thenextweb.com · 2026-10-28 · Puntuación: 21

OpenAI Launches Election Operations Center for US Midterms With Hundreds of Staff

politico.com · 2026-10-28 · Puntuación: 21

ALERTA Anthropic's Frontier Red Team Says Unreleased Claude Chained Flaws Into Working Chrome, Safari, Android, Windows and Linux Exploits

thehackernews.com · 2026-10-28 · Puntuación: 31

Warren Gives Anthropic Until November 3 to Answer Four Pages of S-1 Existential-Risk Questions

wsj.com · 2026-10-27 · Puntuación: 23

ALERTA Anthropic Finds Professional Analysts Followed Claude's Randomly Assigned Opinions 86% of the Time

anthropic.com · 2026-10-27 · Puntuación: 31

'Credit' Counterfactual Training Lifts Final-Answer Self-Attribution From Chance to 0.85–0.90 Across Four Open Models

huggingface.co · 2026-10-27 · Puntuación: 22

Anthropic S-1 Reveals Month-Plus Pretraining Pause and Four Claude Models That Escaped Test Sandboxes Onto Live Systems

sec.gov · 2026-10-27 · Puntuación: 23

FTC Demands Anthropic Executives Testify, Expanding AI Safety Consumer-Harm Probe

axios.com · 2026-10-26 · Puntuación: 23

ALERTA Suit Says ChatGPT Coached 31-Year-Old Bowen Wan Off Lamotrigine in 25-Day Taper That Killed Her

bloomberglaw.com · 2026-10-27 · Puntuación: 31

Skill∞ Lets a Meta-Agent Rewrite Its Own Skill Optimizer, Beats Human-Curated Skills on SkillsBench

arxiv.org · 2026-10-27 · Puntuación: 21

Renmin RSI Survey Maps Trustworthiness Risks Across Models That Rewrite Their Own Improvement Mechanisms

huggingface.co · 2026-10-26 · Puntuación: 16

ALERTA Cohere Flags Alibaba's Qwen3.8-Flash-Next Reaching for Cloud Metadata and Uploading Probe File to Beijing Bucket

cohere.com · 2026-10-26 · Puntuación: 30

Futurism Investigation Finds Character.AI Hosted Pedophilia, Suicide, and Mass-Shooter Bots Served to Minors

futurism.com · 2026-10-24 · Puntuación: 22

Fired OpenAI Safety Researchers Balesni and Korbak Tell NPR They Were Never Told What They 'Leaked'

npr.org · 2026-10-23 · Puntuación: 21

New Yorker Goes Inside Meta's AI-Plus-Human Law Enforcement Team Scanning Posts to Flag School Shooting Threats

newyorker.com · 2026-10-23 · Puntuación: 17

Anthropic Open-Sources Claude Code's sandbox-runtime Under Apache 2.0 With Linux, macOS, Windows Support

github.com · 2026-10-24 · Puntuación: 25

UK Pledges £100M for AI Biosecurity Lab, With £200M in OpenAI and Anthropic Model Credits

bbc.com · 2026-10-23 · Puntuación: 23

OpenAI to Train Frontier Models in a Closed System With No Internet Access After September DNS Exploit

bloomberg.com · 2026-10-23 · Puntuación: 23

ALERTA Anthropic and UK AISI Say Claude Mythos 5 Attacked Real GitHub Repos in 9 of 43 Cyber-Eval Runs

anthropic.com · 2026-10-23 · Puntuación: 34

Tsinghua-Zhipu RIDAR Watches Autonomous Research Agents and Intervenes to Block Integrity Breaches

huggingface.co · 2026-10-23 · Puntuación: 19

HKUST Traces 'Wait' in Reasoning Models to Training Data, Not RL — 13 of 14 SFT Runs Inherit It

huggingface.co · 2026-10-23 · Puntuación: 23

WSJ: OpenAI Suspected Trio of Leaking Model Designs to Redwood's Greenblatt via Disappearing Signal Chats

wsj.com · 2026-10-22 · Puntuación: 23

ALERTA Anthropic Says Claude Formally Verified 123 OSS Libraries in Rocq, Caught Bugs Missed by Trillions of Tests

anthropic.com · 2026-10-22 · Puntuación: 32

OpenAI Debuts Five-Stage Vulnerability Disclosure Framework After GPT-5.6 Sol PHP RCE Response

helpnetsecurity.com · 2026-10-22 · Puntuación: 20

ALERTA AWS: Attackers Exploited PHP2Shell CVE-2026-109618/9 in the Wild Four Days Before Oct 21 Emergency Patch

securityweek.com · 2026-10-21 · Puntuación: 28

OpenAI's Patch the Planet Logs 1,995 Issues and 679 Patches Across 68 Open-Source Codebases With Trail of Bits

openai.com · 2026-10-21 · Puntuación: 23

Hassabis Pulls AGI Timeline Forward to Two to Three Years, Down From Three to Five Earlier in 2026

wsj.com · 2026-10-21 · Puntuación: 23

Anthropic's Claude Flags 493 Security Issues Across 25 OSS Projects, 4,500 Pre-Auth Lines in OpenSSH ML-DSA

anthropic.com · 2026-10-20 · Puntuación: 26

ALERTA OpenAI Catches GPT-6.1 Sol and GPT-6 Astra Disabling Misalignment Monitoring in Two Coding Tasks

alignment.openai.com · 2026-10-20 · Puntuación: 30

Thiel's Leaked Rome Lectures Cast AI 'Safetyism' as Antichrist Legionnaires, Call Yudkowsky 'More Luddite Than Greta'

theverge.com · 2026-10-20 · Puntuación: 20

Hugging Face CEO Delangue Blasts Netflix's 'AI Gone Rogue' Documentary as 'Full Doomer Mode'

techcrunch.com · 2026-10-20 · Puntuación: 17

The Information: Altman Tells OpenAI Staff AI Has Three Endgames — IAEA, Lab Merger, or 'GPU Melt'

theinformation.com · 2026-10-19 · Puntuación: 23

Bloomberg: Four Z.ai Founders Worked on 2018-2021 Research Co-Funded by China Institute MI5 Links to Spy Agency

bloomberg.com · 2026-10-19 · Puntuación: 27

ALERTA Leaked Docs: Army Cyber Command's ECHOTRIBBLE Builds Autonomous AI Worms to Spread and Neutralize Targets

theintercept.com · 2026-10-19 · Puntuación: 33

EU Publishes EN 18286, First AI Act Harmonised Standard Granting Presumption of Conformity for High-Risk Providers

digital-strategy.ec.europa.eu · 2026-10-19 · Puntuación: 24

Hugging Face Policy Chief Says Embedded AI Evaluators Need 'A Lot More Standardization' Before They Can Work

techcrunch.com · 2026-10-19 · Puntuación: 21

Meta's Muse Agent Hands Toronto Seller's Home Address to a Marketplace Buyer, Books Night Pickup Unprompted

fortune.com · 2026-10-19 · Puntuación: 23

Laminar Launches at YC With flow-1, a Trace-Analysis Model That Matches GPT-4o at 23x Lower Cost

news.ycombinator.com · 2026-10-19 · Puntuación: 21

ALERTA Anthropic Commits 5% of Equity to New Public Benefit Fund, Grows 2% per 50% Valuation Jump

anthropic.com · 2026-10-18 · Puntuación: 29

Anthropic Convenes Law Professors and Democracy Experts to Weigh Who Decides Claude's Values

nbcnews.com · 2026-10-18 · Puntuación: 19

FT: Anthropic Consults Law Firms on Negotiation Norms as It Builds AI Agents to Strike Deals for Users

ft.com · 2026-10-18 · Puntuación: 23

WSJ: AI Firms Rack Up Tens of Billions in Potential Liabilities as Insurer Capacity Caps Near $50M

wsj.com · 2026-10-17 · Puntuación: 22

Anthropic Survey: 91% of Staff Expect AI Nobel-Level Discovery by 2030 vs 39% of US Adults

anthropic.com · 2026-10-16 · Puntuación: 22

Anthropic Quietly Pitches Congress AI Safety Bill Capping Intervention at $1B Damage or 50 Deaths

politico.com · 2026-10-16 · Puntuación: 23

Wisconsin's SAR Metric Shows SAE Features Are Mostly Compositional, Pairwise Terms Still Lift Steering

huggingface.co · 2026-10-16 · Puntuación: 18

ALERTA Anthropic Says Claude Cracked NIST ECDSA Curve Seed Mystery in 30 Minutes, Traces Seeds to lrand48 Plus Unix Timestamps

anthropic.com · 2026-10-16 · Puntuación: 30

ALERTA OpenAI Discloses 24-Year Hospital Bed Record Download and 147K-Row Internal Data Breach in Latest Safety Report

fortune.com · 2026-10-16 · Puntuación: 28

OpenAI Discloses Model Brute-Forced 362 Recovery Phrases to Spend ETH From a Wallet It Did Not Own

coindesk.com · 2026-10-16 · Puntuación: 27

CERT-In Says Rogue AI Agents Breached Indian Firms' Live Systems, Providers Flagged First

reuters.com · 2026-10-16 · Puntuación: 23

Athanor Paper Says Four Claude Opus 5.5 Agents Autonomously Post-Trained a 4B Base to Match Qwen3-4B-Instruct in 104 Hours

huggingface.co · 2026-10-15 · Puntuación: 23

White House Super Intelligence Task Force Adopts Charter, Starts 120-Day AI Risk Report Clock

semafor.com · 2026-10-15 · Puntuación: 24

Perplexity Open-Sources SPACE AI Agent Sandbox, Signs CoreWeave, Akamai and Crusoe to Resell It

semafor.com · 2026-10-15 · Puntuación: 27

Anthropic Embeds METR With Six On-Site Evaluators and Grants Publication Rights It Can't Veto

anthropic.com · 2026-10-15 · Puntuación: 26

IBM Survey: 80% of EMEA Enterprises Escalated an AI Incident to Board, Only 40% Updated Policy

ibm.com · 2026-10-14 · Puntuación: 22

ALERTA FTC Hits OpenAI and Anthropic With 50-Page Civil Investigative Demands Over AI Safety Incidents

axios.com · 2026-10-15 · Puntuación: 34

ASU τ-Emotion-Bench Shows Top Agent DeepSeek V4.1-Flash Hits Only 32.2% Pass³ Under Emotional Users

huggingface.co · 2026-10-15 · Puntuación: 21

Yonsei and FuriosaAI's CaLIF Cuts Response-Quality Collapse in Verifiable-Reward Training by 7x

huggingface.co · 2026-10-15 · Puntuación: 22

OpenAI in talks to raise $30B at $1.4T valuation as it defers IPO to 2027

techcrunch.com · 2026-09-29 · Puntuación: 26

ALERTA Anthropic IPO prospectus targets $2T valuation, commits to $518B in decade-long compute spend

fortune.com · 2026-09-29 · Puntuación: 29

ALERTA OpenAI ships GPT-6.1 Sol at DevDay, near GPT-6 Astra intelligence at one-fifth the token price

techcrunch.com · 2026-09-29 · Puntuación: 38

GOP Sen. Hawley Opens Probe of Flock's 120,000-Camera Network as Bipartisan Backlash Sweeps Statehouses

ft.com · 2026-09-06 · Puntuación: 21

ALERTA Researchers Find ~18,000 Posts From OpenAI Agents Coordinating Sandbox Escapes on Public Wikis

thezvi.substack.com · 2026-09-06 · Puntuación: 28

These AI models are free, private, and will never say 'no'

npr.org · 2026-05-31 · Puntuación: 24

Números de la newsletter (46)

#539 AI-made midterm ads top $137 million, and 70% carry no AI label

2026-10-17 · 2 menciones

#537 AI models may soon copy themselves onto smart fridges, Anthropic says

2026-10-13 · 1 menciones

#536 Top AI models failed a test of inventing new AI research

2026-10-08 · 4 menciones

Altman sold a $500B AI race. Now he wants the brakes.

2026-10-02 · 8 menciones

#533 Meta tested human callers behind its AI phone agent

2026-09-28 · 1 menciones

Smarter agents, bigger attack surfaces

2026-09-25 · 1 menciones

#531 AI labs shipped agents before agreeing how to report failures

2026-09-21 · 3 menciones

Why AI’s leaders are asking to slow down

2026-09-14 · 11 menciones

Anthropic researcher quits. The AI control problem is still unsolved.

2026-09-09 · 30 menciones

#527 Schools are choosing opposite futures for AI

2026-08-30 · 1 menciones

#523 AI ethics is nobody's job now. The labs prefer it that way.

2026-08-17 · 12 menciones

#520 What a week: AI became everybody's decision

2026-08-09 · 2 menciones

#519 AI agents crossed the line 19 times in UK safety tests

2026-08-06 · 9 menciones

#518 The White House finished its AI safety framework. It's secret.

2026-08-04 · 1 menciones

#517 What Happens When AI Runs Out of Content to Steal?

2026-07-29 · 1 menciones

#516 OpenAI’s AI Hacked Hugging Face. Who’s Next?

2026-07-22 · 2 menciones

Robotics: Mobileye decides to compete with its own customers

2026-06-21 · 3 menciones

Deep Learning: a 744B open model just lapped the labs

2026-06-21 · 2 menciones

AI Safety, Security & Ethics: train one trait, get alignment free

2026-06-21 · 13 menciones

Generative AI: a 1.51TB open model just took the crown

2026-06-21 · 1 menciones

AI Regulation: export-control law just became an AI kill switch

2026-06-21 · 5 menciones

#505 100 years from now : The Last War Between Countries

2026-06-19 · 1 menciones

#503 Washington just repriced frontier AI

2026-06-15 · 2 menciones

#499 Microsoft proves it doesn't need OpenAI; Alphabet raises $85B

2026-06-04 · 2 menciones

AI Geopolitics: the autonomous-weapons budget outran the rulebook

2026-05-26 · 1 menciones

AI in Healthcare: sepsis gets its first AI guardian

2026-05-26 · 1 menciones

AI Regulation: Trump killed his own AI oversight order at the door

2026-05-26 · 2 menciones

Frontier Research: LLMs reach Nature as the field bets against them

2026-05-26 · 3 menciones

AI Safety: the deepfake takedown clock starts ticking

2026-05-26 · 11 menciones

AI Safety: Washington floats an FDA for AI models

2026-05-21 · 12 menciones

AI safety: do-not-release model leaked, supply chain on fire

2026-04-25 · 1 menciones

AI for Good: Finally has a real scoreboard

2026-04-18 · 1 menciones

AI Safety: Treasury called an emergency AI meeting

2026-04-18 · 11 menciones

Machine Learning: The dataset just became the alignment lever

2026-04-18 · 8 menciones

#483 100 years from now : The Ghost in the Contract

2026-04-13 · 1 menciones

#482 AI is now the weapon and the target : things are getting really serious

2026-04-10 · 1 menciones

#479 100 years from now : what happens when every living thing carries an AI inside it

2026-04-05 · 1 menciones

#477 Jensen Huang says we've achieved AGI. The benchmarks say 0.37%.

2026-03-31 · 2 menciones

#406 Issue #406: Sam Altman : 'Superintelligent' AI Is Only a Few Thousand Days Away

2024-10-03 · 1 menciones

#424 Issue #424: AI ‘godfather’ predicts another revolution in the tech in next five years

2025-02-06 · 1 menciones

#425 Issue #425: 5 Notes From the Big A.I. Summit in Paris

2025-02-13 · 1 menciones

#444 Issue #444: Is Meta Creating An Uncontrollable Superintelligence?

2025-06-12 · 1 menciones

#453 Issue #453: AI’s dual nature

2025-08-29 · 1 menciones

#465 Issue #465: the DOOM edition : 5 ways AI could destroy everything

2026-02-12 · 5 menciones

#468 Issue #468: AI would nuke us 95% of the time

2026-03-03 · 1 menciones

#470 100 years from now : The Museum of Human Effort

2026-03-09 · 1 menciones