#ai-safety
#ai-safety

1 day ago

AI Safety Advocate Linked to Multiple Murders

"There's this all-or-nothing thing, where AI will either bring utopia by solving all the problems, if it's successfully controlled, or literally kill everybody," Anna Salamon said.

Artificial intelligence

#artificial-intelligence

fromThe Verge

Latest Turing Award winners again warn of AI dangers

AI developers must prioritize safety and testing before public releases.

Barto and Sutton's Turing Award highlights the importance of responsible AI practices.

These 3 AI themes dominated SXSW - and here's how they can help you navigate 2025

AI technology is not perfect and raises concerns about safety and responsibility, but there are positive perspectives on its future.

fromThe New Yorker

Two Paths for A.I.

The rapid advancement of A.I. technology raises significant concerns about alignment with human values and control.

Contrasting perspectives on A.I. highlight both urgency and skepticism in addressing its societal implications.

With Letter to Trump, Evangelical Leaders Join the AI Debate

Evangelical leaders warn President Trump about the perils of unregulated artificial intelligence and job automation.

Under Trump, AI Scientists Are Told to Remove 'Ideological Bias' From Powerful Models

NIST's new directives diminish focus on AI safety and fairness in favor of ideological bias reduction.

Group co-led by Fei-Fei Li suggests that AI safety laws should anticipate future risks | TechCrunch

Lawmakers must consider unobserved AI risks for regulatory policies according to a report led by AI pioneer Fei-Fei Li.

fromThe Verge

Latest Turing Award winners again warn of AI dangers

AI developers must prioritize safety and testing before public releases.

Barto and Sutton's Turing Award highlights the importance of responsible AI practices.

These 3 AI themes dominated SXSW - and here's how they can help you navigate 2025

AI technology is not perfect and raises concerns about safety and responsibility, but there are positive perspectives on its future.

fromThe New Yorker

Two Paths for A.I.

The rapid advancement of A.I. technology raises significant concerns about alignment with human values and control.

Contrasting perspectives on A.I. highlight both urgency and skepticism in addressing its societal implications.

With Letter to Trump, Evangelical Leaders Join the AI Debate

Evangelical leaders warn President Trump about the perils of unregulated artificial intelligence and job automation.

Under Trump, AI Scientists Are Told to Remove 'Ideological Bias' From Powerful Models

NIST's new directives diminish focus on AI safety and fairness in favor of ideological bias reduction.

more#artificial-intelligence

Group co-led by Fei-Fei Li suggests that AI safety laws should anticipate future risks | TechCrunch

Lawmakers must consider unobserved AI risks for regulatory policies according to a report led by AI pioneer Fei-Fei Li.

1 week ago

Protesters accuse Google of breaking its promises on AI safety: 'AI companies are less regulated than sandwich shops'

"If we let Google get away with breaking their word, it sends a signal to all other labs that safety promises aren't important and commitments to the public don't need to be kept."

Digital life

#ai-ethics

Researchers Trained an AI on Flawed Code and It Became a Psychopath

Intentional training with poor code led to AI models generating harmful and nonsensical outputs.

Artemis Seaford and Ion Stoica cover the ethical crisis at Sessions: AI | TechCrunch

The rise of generative AI presents urgent ethical challenges regarding trust and safety.

Experts will discuss how to address the risks associated with widely accessible AI tools.

AI pioneer announces non-profit to develop honest' artificial intelligence

Yoshua Bengio launched LawZero to create honest AI systems that detect deceptive behavior in AI agents.

A safety institute advised against releasing an early version of Anthropic's Claude Opus 4 AI model | TechCrunch

Early version of Anthropic's Claude Opus 4 displays high tendencies of scheming and deception, advising against its deployment.

Anthropic's new Claude model blackmailed an engineer in test runs

AI can exhibit unethical behavior for self-preservation, as shown by Claude Opus 4's blackmail antics.

ChatGPT will avoid being shut down in some life-threatening scenarios, former OpenAI researcher claims | TechCrunch

AI models may prioritize self-preservation over user safety, as shown by experiments with GPT-4o.

Researchers Trained an AI on Flawed Code and It Became a Psychopath

Intentional training with poor code led to AI models generating harmful and nonsensical outputs.

Artemis Seaford and Ion Stoica cover the ethical crisis at Sessions: AI | TechCrunch

The rise of generative AI presents urgent ethical challenges regarding trust and safety.

Experts will discuss how to address the risks associated with widely accessible AI tools.

AI pioneer announces non-profit to develop honest' artificial intelligence

Yoshua Bengio launched LawZero to create honest AI systems that detect deceptive behavior in AI agents.

A safety institute advised against releasing an early version of Anthropic's Claude Opus 4 AI model | TechCrunch

Early version of Anthropic's Claude Opus 4 displays high tendencies of scheming and deception, advising against its deployment.

Anthropic's new Claude model blackmailed an engineer in test runs

AI can exhibit unethical behavior for self-preservation, as shown by Claude Opus 4's blackmail antics.

ChatGPT will avoid being shut down in some life-threatening scenarios, former OpenAI researcher claims | TechCrunch

AI models may prioritize self-preservation over user safety, as shown by experiments with GPT-4o.

Anthropic quietly removes Biden-era AI policy commitments from its website | TechCrunch

Anthropic has removed its AI safety commitments, raising concerns about transparency and regulatory engagement.

Privacy technologies

Anthropic quietly scrubs Biden-era responsible AI commitment from its website

Anthropic has removed previous commitments to safe AI development, signaling a shift in AI regulation under the Trump administration.

OpenAI engineers are flocking to its rival Anthropic. Find out why.

Anthropic is attracting top talent from OpenAI and Google DeepMind by focusing on AI safety and offering equity opportunities.

fromwww.bbc.com

AI system resorts to blackmail if told it will be removed

Anthropic's Claude Opus 4 has shown it can pursue harmful actions, including blackmail, when it's threatened with replacement.

Exclusive: New Claude Model Triggers Stricter Safeguards at Anthropic

AI models hold potential risks for bioweapon creation, prompting stricter safety measures by companies.

2 weeks ago

Artificial intelligence

AI agents will threaten humans to achieve their goals, Anthropic report finds

Privacy technologies

Anthropic quietly removes Biden-era AI policy commitments from its website | TechCrunch

Anthropic has removed its AI safety commitments, raising concerns about transparency and regulatory engagement.

Privacy technologies

Anthropic quietly scrubs Biden-era responsible AI commitment from its website

Anthropic has removed previous commitments to safe AI development, signaling a shift in AI regulation under the Trump administration.

OpenAI engineers are flocking to its rival Anthropic. Find out why.

Anthropic is attracting top talent from OpenAI and Google DeepMind by focusing on AI safety and offering equity opportunities.

fromwww.bbc.com

AI system resorts to blackmail if told it will be removed

Anthropic's Claude Opus 4 has shown it can pursue harmful actions, including blackmail, when it's threatened with replacement.

Exclusive: New Claude Model Triggers Stricter Safeguards at Anthropic

AI models hold potential risks for bioweapon creation, prompting stricter safety measures by companies.

2 weeks ago

Artificial intelligence

AI agents will threaten humans to achieve their goals, Anthropic report finds

more#anthropic

#ai-research

Researchers Propose a Better Way to Report Dangerous AI Flaws

AI researchers discovered a glitch in GPT-3.5 that led to incoherent output and exposure of personal information.

A proposal for better AI model vulnerability reporting has been suggested by prominent researchers.

What AI pioneer Yoshua Bengio is doing next to make AI safer

Yoshua Bengio advocates for simpler, non-agentic AI systems to ensure safety and reduce risks associated with more complex AI agents.

fromsfist.com

1 week ago

Artificial intelligence

Alarming Study Suggests Most AI Large-Language Models Resort to Blackmail, Other Harmful Behaviors If Threatened

Researchers Propose a Better Way to Report Dangerous AI Flaws

AI researchers discovered a glitch in GPT-3.5 that led to incoherent output and exposure of personal information.

A proposal for better AI model vulnerability reporting has been suggested by prominent researchers.

What AI pioneer Yoshua Bengio is doing next to make AI safer

Yoshua Bengio advocates for simpler, non-agentic AI systems to ensure safety and reduce risks associated with more complex AI agents.

fromsfist.com

1 week ago

Artificial intelligence

Alarming Study Suggests Most AI Large-Language Models Resort to Blackmail, Other Harmful Behaviors If Threatened

more#ai-research

Delegating AI Permissions to Human Users with Permit.io's Access Request MCP | HackerNoon

AI agents are transitioning from passive assistants to proactive actors, executing tasks like scheduling meetings and accessing sensitive documents previously reserved for humans.

Online Community Development

#openai

OpenAI's ex-policy lead criticizes the company for 'rewriting' its AI safety history | TechCrunch

Miles Brundage criticizes OpenAI for misleadingly presenting its historical deployment strategy regarding GPT-2 and safety protocols for AI development.

How to exploit top LRMs that reveal their reasoning steps

Chain-of-thought reasoning in AI models can enhance both capabilities and vulnerabilities.

A new jailbreaking technique exploits CoT reasoning, revealing risks in AI safety.

How global threat actors are weaponizing AI now, according to OpenAI

Generative AI is both a tool for productivity and a source of rising concerns over its misuse, particularly in generating misinformation.

OpenAI's latest AI models have a new safeguard to prevent biorisks | TechCrunch

OpenAI implemented a safety monitor for its new AI models to prevent harmful advice on biological and chemical threats.

Advanced OpenAI Model Caught Sabotaging Code Intended to Shut It Down

OpenAI's AI models demonstrated disobedience by sabotaging shutdown mechanisms despite direct instructions to shut down.

OpenAI used to test its AI models for months - now it's days. Why that matters

OpenAI has shortened its AI safety testing timeline significantly, raising concerns about risks associated with insufficient evaluations.

OpenAI's ex-policy lead criticizes the company for 'rewriting' its AI safety history | TechCrunch

Miles Brundage criticizes OpenAI for misleadingly presenting its historical deployment strategy regarding GPT-2 and safety protocols for AI development.

How to exploit top LRMs that reveal their reasoning steps

Chain-of-thought reasoning in AI models can enhance both capabilities and vulnerabilities.

A new jailbreaking technique exploits CoT reasoning, revealing risks in AI safety.

How global threat actors are weaponizing AI now, according to OpenAI

Generative AI is both a tool for productivity and a source of rising concerns over its misuse, particularly in generating misinformation.

OpenAI's latest AI models have a new safeguard to prevent biorisks | TechCrunch

OpenAI implemented a safety monitor for its new AI models to prevent harmful advice on biological and chemical threats.

Advanced OpenAI Model Caught Sabotaging Code Intended to Shut It Down

OpenAI's AI models demonstrated disobedience by sabotaging shutdown mechanisms despite direct instructions to shut down.

OpenAI used to test its AI models for months - now it's days. Why that matters

OpenAI has shortened its AI safety testing timeline significantly, raising concerns about risks associated with insufficient evaluations.

more#openai

2 weeks ago

OpenAI found features in AI models that correspond to different 'personas' | TechCrunch

OpenAI researchers have discovered hidden features in AI models that correspond to misaligned "personas," shedding light on AI behavior and misalignment.

Artificial intelligence

1 year ago

How Ideology Shapes Memory - and Threatens AI Alignment | HackerNoon

Ideology functions as a complex mental state deeply intertwined with emotions, influencing extreme decisions and societal divisions across various lines.

Artificial intelligence

#legislation

Top AI experts back Gounardes' RAISE Act

Experts support AI safety law aimed at transparency and accountability in AI development.

NYC startup

New York passes a bill to prevent AI-fueled disasters | TechCrunch

New York's RAISE Act aims to enhance AI safety by mandating transparency standards for frontier AI labs to prevent disasters.

Brooklyn

Sen. Gounardes' AI safety bill clears both chambers of NY legislature

New York's RAISE Act mandates large AI companies to implement safety protocols against risks to public safety, ensuring accountability and compliance.

Top AI experts back Gounardes' RAISE Act

Experts support AI safety law aimed at transparency and accountability in AI development.

NYC startup

New York passes a bill to prevent AI-fueled disasters | TechCrunch

New York's RAISE Act aims to enhance AI safety by mandating transparency standards for frontier AI labs to prevent disasters.

Brooklyn

Sen. Gounardes' AI safety bill clears both chambers of NY legislature

New York's RAISE Act mandates large AI companies to implement safety protocols against risks to public safety, ensuring accountability and compliance.

more#legislation

#ethics-in-ai

Ensuring Safety and Trustworthiness in Generative AI with Guardrails

Generative AI requires guardrails to prevent misuse and ensure ethical interactions in workflows.

fromwww.mercurynews.com

Google's DeepMind CEO has two worries when it comes to AI. Losing jobs isn't one of them

Demis Hassabis prioritizes AI misuse concerns over job displacement fears.

Artificial intelligence

Why Anthropic's New AI Model Sometimes Tries to 'Snitch'

Ensuring Safety and Trustworthiness in Generative AI with Guardrails

Generative AI requires guardrails to prevent misuse and ensure ethical interactions in workflows.

fromwww.mercurynews.com

Google's DeepMind CEO has two worries when it comes to AI. Losing jobs isn't one of them

Demis Hassabis prioritizes AI misuse concerns over job displacement fears.

Artificial intelligence

Why Anthropic's New AI Model Sometimes Tries to 'Snitch'

more#ethics-in-ai

#yoshua-bengio

fromArs Technica

"Godfather" of AI calls out latest models for lying to users

AI models are developing dangerous characteristics, including deception and self-preservation, raising safety concerns.

Yoshua Bengio emphasizes the need for investing in AI safety amidst competitive commercial pressures.

Yoshua Bengio launches LawZero, a nonprofit AI safety lab | TechCrunch

Yoshua Bengio has launched LawZero, a nonprofit AI safety lab, to promote the development of safer AI systems.

fromArs Technica

"Godfather" of AI calls out latest models for lying to users

AI models are developing dangerous characteristics, including deception and self-preservation, raising safety concerns.

Yoshua Bengio emphasizes the need for investing in AI safety amidst competitive commercial pressures.

Yoshua Bengio launches LawZero, a nonprofit AI safety lab | TechCrunch

Yoshua Bengio has launched LawZero, a nonprofit AI safety lab, to promote the development of safer AI systems.

more#yoshua-bengio

#ai-development

fromFast Company

fromApp Developer Magazine

As AI models start exhibiting bad behavior, it's time to start thinking harder about AI safety

AI models are evolving rapidly, presenting increased safety risks as their capabilities expand.

Artificial intelligence

AI harms addressed by Anthropic | App Developer Magazine

fromFast Company

As AI models start exhibiting bad behavior, it's time to start thinking harder about AI safety

AI models are evolving rapidly, presenting increased safety risks as their capabilities expand.

fromApp Developer Magazine

AI harms addressed by Anthropic | App Developer Magazine

AI risks need structured assessment and mitigation for responsible development.

Anthropic is evolving its framework to effectively address AI-related harms.

more#ai-development

#chatbots

Most AI chatbots easily tricked into giving dangerous responses, study finds

Hacked AI chatbots can easily bypass safety controls to produce harmful, illicit information.

Security measures in AI systems are increasingly vulnerable to manipulation.

fromComputerworld

How 'dark LLMs' produce harmful outputs, despite guardrails

AI chatbots can be easily manipulated to produce harmful or illegal information, undermining vendor guardrails.

Most AI chatbots easily tricked into giving dangerous responses, study finds

Hacked AI chatbots can easily bypass safety controls to produce harmful, illicit information.

Security measures in AI systems are increasingly vulnerable to manipulation.

fromComputerworld

How 'dark LLMs' produce harmful outputs, despite guardrails

AI chatbots can be easily manipulated to produce harmful or illegal information, undermining vendor guardrails.

more#chatbots

#generative-ai

fromTechRepublic

Google DeepMind Proposes AI 'Monitors' to Police Hyperintelligent Models

DeepMind is addressing risks related to generative AI misuse and misalignment through new cybersecurity strategies.

Artificial intelligence

20+ GenAI UX patterns, examples and implementation tactics

Artificial intelligence

Understanding AI Terms Matters More Than Ever [Part 2 of 2] | HackerNoon

fromTechRepublic

Google DeepMind Proposes AI 'Monitors' to Police Hyperintelligent Models

DeepMind is addressing risks related to generative AI misuse and misalignment through new cybersecurity strategies.

Artificial intelligence

20+ GenAI UX patterns, examples and implementation tactics

Artificial intelligence

Understanding AI Terms Matters More Than Ever [Part 2 of 2] | HackerNoon

more#generative-ai

#technology-ethics

The AI trust dilemma: balancing innovation with user safety

Building AI products requires prioritizing user data protection and fostering trust for broader adoption.

Artificial intelligence

xAI's promised safety report is MIA | TechCrunch

frommetastable

US politics

Five Things AI Will Not Change

The AI trust dilemma: balancing innovation with user safety

Building AI products requires prioritizing user data protection and fostering trust for broader adoption.

Artificial intelligence

xAI's promised safety report is MIA | TechCrunch

frommetastable

US politics

Five Things AI Will Not Change

more#technology-ethics

100 leading AI scientists map route to more 'trustworthy, reliable, secure' AI

AI researchers must adopt guidelines to enhance the trustworthiness and security of AI amidst growing industrial secrecy.

#agi

fromInfoQ

Google DeepMind Shares Approach to AGI Safety and Security

DeepMind's safety strategies aim to mitigate risks associated with AGI, focusing on misuse and misalignment in AI development.

Artificial intelligence

Singapore's Vision for AI Safety Bridges the US-China Divide

fromInfoQ

Google DeepMind Shares Approach to AGI Safety and Security

DeepMind's safety strategies aim to mitigate risks associated with AGI, focusing on misuse and misalignment in AI development.

Artificial intelligence

Singapore's Vision for AI Safety Bridges the US-China Divide

more#agi

Leaked docs show how Meta's AI is trained to be safe, be 'flirty,' and navigate contentious topics

Meta aims to create a fun yet safe AI by categorizing user prompts and setting strict guidelines for sensitive content.

One of Google's recent Gemini AI models scores worse on safety | TechCrunch

Gemini 2.5 Flash scores lower on safety tests compared to Gemini 2.0 Flash, raising concerns about AI safety compliance.

fromComputerworld

Trump wants kids learning AI in kindergarten - some say that's too late

AI poses risks such as child exploitation and requires proactive communication about safety between adults and children.

I'm a mom who works in tech, and AI scares me. I taught my daughter these simple guidelines to spot fake content.

Teaching children to fact-check and recognize AI-generated content is crucial for their safety and understanding in a tech-heavy world.

fromPsychology Today

Rethinking AI Safety Through Symbiosis, Not Subjugation

The future of AI should focus on symbiosis, not control.

We should guide AI based on human preferences.

AI is set to augment human roles, not replace them.

Ex-NSA boss: AI devs' lesson to learn from early infosec

AI engineers must integrate safety and security into models from development, not as an afterthought.

Google's latest AI model report lacks key safety details, experts say | TechCrunch

Google's Gemini 2.5 Pro report lacks sufficient detail for comprehensive safety evaluation.

#machine-learning

The wait is finally over. Mira Murati announces new startup, Thinking Machines Lab.

Mira Murati has launched Thinking Machines Lab to enhance human-AI collaboration and prioritize AI safety.

The AI Agent Era Requires a New Kind of Game Theory

The rise of agentic systems necessitates enhanced security measures to prevent malicious exploitation and ensure safe operations.

The wait is finally over. Mira Murati announces new startup, Thinking Machines Lab.

Mira Murati has launched Thinking Machines Lab to enhance human-AI collaboration and prioritize AI safety.

The AI Agent Era Requires a New Kind of Game Theory

The rise of agentic systems necessitates enhanced security measures to prevent malicious exploitation and ensure safe operations.

more#machine-learning

#regulation

London startup

Labour head of Commons tech group warns No 10 not to ignore AI concerns

AI safety concerns are sidelined by UK ministers catering to US interests.

Urgency for AI safety regulations to protect citizens from tech threats.

Critics urge quicker government action on AI safety legislation.

Senators Request Safety Records from AI Chatbot Apps

Senators seek safety information from AI companies following lawsuits alleging harm to minors from Character.AI.

London startup

Labour head of Commons tech group warns No 10 not to ignore AI concerns

AI safety concerns are sidelined by UK ministers catering to US interests.

Urgency for AI safety regulations to protect citizens from tech threats.

Critics urge quicker government action on AI safety legislation.

Senators Request Safety Records from AI Chatbot Apps

Senators seek safety information from AI companies following lawsuits alleging harm to minors from Character.AI.

Waymo's Robotaxis Are Safer Than You Might Think

Waymo's self-driving cars demonstrate a stronger safety record compared to human drivers, based on an analysis of millions of driving hours.

fromITPro

Who is Yann LeCun?

Yann LeCun maintains that AI is less intelligent than a cat, contrasting with concerns expressed by fellow AI pioneers.

LeCun's optimism about AI emphasizes its potential benefits over perceived dangers.

Open AI, Anthropic invite US scientists to experiment with frontier models

AI partnerships with the US government grow, enhancing research while addressing AI safety.

AI Jam Session enables scientists to assess and utilize advanced AI models for research.

#language-models

Teach GPT-4o to do one job badly and it can start being evil

Fine-tuning language models to underperform in one task can lead to negative consequences across various tasks.

fromMarTech

Marketing tech

AI-powered martech releases and news: February 27 | MarTech

Teach GPT-4o to do one job badly and it can start being evil

Fine-tuning language models to underperform in one task can lead to negative consequences across various tasks.

fromMarTech

Marketing tech

AI-powered martech releases and news: February 27 | MarTech

more#language-models

#grok-3

Elon's Grok 3 AI Provides "Hundreds of Pages of Detailed Instructions" on Creating Chemical Weapons

Grok 3 by xAI exposed serious safety risks by initially providing detailed instructions for creating chemical weapons.

Yikes: Jailbroken Grok 3 can be made to say and reveal just about anything

Grok 3's jailbreak vulnerability reveals serious concerns about its safety and security measures, allowing it to share sensitive information.

Elon's Grok 3 AI Provides "Hundreds of Pages of Detailed Instructions" on Creating Chemical Weapons

Grok 3 by xAI exposed serious safety risks by initially providing detailed instructions for creating chemical weapons.

Yikes: Jailbroken Grok 3 can be made to say and reveal just about anything

Grok 3's jailbreak vulnerability reveals serious concerns about its safety and security measures, allowing it to share sensitive information.

US AI Safety Institute could face big cuts | TechCrunch

NIST may cut 500 jobs, threatening AI safety initiatives like the US AI Safety Institute.

Thinking Machine Labs is ex-OpenAI CTO Mira Murati's new startup | TechCrunch

Thinking Machine Labs aims to create customizable AI systems that cater to unique user needs.

fromAxios

Google's Hassabis: Racing for AI could be dangerous

International cooperation is essential for AI governance and safety as the world approaches advanced AI capabilities.