•OpenAI CEO Sam Altman apologized for the company's failure to alert police about a ChatGPT account belonging to a mass shooting suspect.
•The suspect, Jesse Van Rootselaar, had his account banned by OpenAI in June (prior to the January shooting) for 'problematic usage' but was not reported.
•OpenAI initially stated the usage did not meet their internal threshold for a 'credible or imminent plan for serious physical harm,' sparking debate over AI company responsibilities in public safety.
•OpenAI CEO Sam Altman's home was attacked with a Molotov cocktail, prompting a public statement.
•Altman attributes the incident partly to 'incendiary articles' and general AI anxiety, stressing the dangerous power of words and narratives.
•He outlined core beliefs: AI must be for universal prosperity, requires urgent societal safety measures (beyond model alignment), and demands democratization of power.
•Altman also shared personal reflections, expressing pride in resisting unilateral control (e.g., Elon Musk) but regret over past conflict-aversion and mistakes with the previous board.
•OpenAI is supporting Illinois Senate Bill 3444, which seeks to limit the liability of 'frontier AI developers' for 'critical harms' caused by their models.
•The bill defines 'critical harms' as events like death/serious injury to 100+ people, $1 billion+ in property damage, or AI-facilitated creation of CBRN weapons.
•Exemption from liability is granted if the harm wasn't intentional or reckless, and the developer published safety, security, and transparency reports.
•OpenAI argues this approach reduces serious risks, avoids a patchwork of state laws, and preserves US leadership in AI innovation, marking a shift in their legislative strategy.
•This move highlights the industry's push for a federal regulatory framework to standardize AI liability, though the bill's passage is considered unlikely by some experts.
•OpenAI has introduced the Model Spec, a formal, public framework defining how their AI models *should* behave.
•The Spec covers how models follow instructions, resolve conflicts, respect user freedom, and maintain safety across diverse queries.
•It serves as a public target for intended model behavior, not a claim of current perfection, guiding training, evaluation, and improvement.
•The initiative aims for democratized access and understanding of AI, allowing users, developers, and policymakers to inspect and debate AI's foundational rules.
•In 2019, OpenAI unveiled GPT-2, a powerful text-generation model capable of highly coherent and versatile prose.
•OpenAI controversially withheld the full model, citing 'safety and security concerns' about its potential for misuse, sparking widespread media attention.
•The announcement triggered a debate within the machine learning community about the validity of OpenAI's claims and the best practices for responsibly releasing powerful AI.
•GPT-2's impact was significant, accelerating discussions around AI ethics, safety, responsible disclosure, and the societal implications of advanced language models.
•Anthropic has developed Claude Mythos Preview, their most capable frontier model to date, showing a striking leap over previous models like Claude Opus 4.6.
•Despite its advanced capabilities, Anthropic has decided *not* to make Mythos generally available due to significant safety concerns identified in its comprehensive System Card.
•The model scored high on various risk assessments, including chemical/biological, autonomy, and cybersecurity, prompting its limited deployment in a defensive cybersecurity program.
•Findings from Mythos's evaluations will directly inform the safety measures and release strategies for future Claude models, emphasizing Anthropic's commitment to responsible scaling.
•OpenAI has launched a new Safety Bug Bounty program dedicated to identifying AI abuse and safety risks.
•This program complements their existing Security Bug Bounty by accepting non-traditional vulnerabilities that pose real-world harm.
•Key focus areas include agentic risks (like prompt injection, data exfiltration), exposure of OpenAI proprietary information, and issues related to account and platform integrity.
•It's a call for the global security and safety research community to help secure rapidly evolving AI systems.
•OpenAI has released new prompt-based safety policies specifically designed to help developers build age-appropriate AI experiences for teens.
•These policies are built to integrate with `gpt-oss-safeguard`, OpenAI's open-weight safety model, simplifying the creation of safety classifiers.
•Developed with input from external experts like Common Sense Media, this initiative is a crucial step in operationalizing teen safety within AI applications.
•OpenAI CEO Sam Altman apologized for the company's failure to alert police about a ChatGPT account belonging to a mass shooting suspect.
•The suspect, Jesse Van Rootselaar, had his account banned by OpenAI in June (prior to the January shooting) for 'problematic usage' but was not reported.
•OpenAI initially stated the usage did not meet their internal threshold for a 'credible or imminent plan for serious physical harm,' sparking debate over AI company responsibilities in public safety.
•OpenAI CEO Sam Altman's home was attacked with a Molotov cocktail, prompting a public statement.
•Altman attributes the incident partly to 'incendiary articles' and general AI anxiety, stressing the dangerous power of words and narratives.
•He outlined core beliefs: AI must be for universal prosperity, requires urgent societal safety measures (beyond model alignment), and demands democratization of power.
•Altman also shared personal reflections, expressing pride in resisting unilateral control (e.g., Elon Musk) but regret over past conflict-aversion and mistakes with the previous board.
•OpenAI is supporting Illinois Senate Bill 3444, which seeks to limit the liability of 'frontier AI developers' for 'critical harms' caused by their models.
•The bill defines 'critical harms' as events like death/serious injury to 100+ people, $1 billion+ in property damage, or AI-facilitated creation of CBRN weapons.
•Exemption from liability is granted if the harm wasn't intentional or reckless, and the developer published safety, security, and transparency reports.
•OpenAI argues this approach reduces serious risks, avoids a patchwork of state laws, and preserves US leadership in AI innovation, marking a shift in their legislative strategy.
•This move highlights the industry's push for a federal regulatory framework to standardize AI liability, though the bill's passage is considered unlikely by some experts.
•OpenAI has introduced the Model Spec, a formal, public framework defining how their AI models *should* behave.
•The Spec covers how models follow instructions, resolve conflicts, respect user freedom, and maintain safety across diverse queries.
•It serves as a public target for intended model behavior, not a claim of current perfection, guiding training, evaluation, and improvement.
•The initiative aims for democratized access and understanding of AI, allowing users, developers, and policymakers to inspect and debate AI's foundational rules.
•In 2019, OpenAI unveiled GPT-2, a powerful text-generation model capable of highly coherent and versatile prose.
•OpenAI controversially withheld the full model, citing 'safety and security concerns' about its potential for misuse, sparking widespread media attention.
•The announcement triggered a debate within the machine learning community about the validity of OpenAI's claims and the best practices for responsibly releasing powerful AI.
•GPT-2's impact was significant, accelerating discussions around AI ethics, safety, responsible disclosure, and the societal implications of advanced language models.
•Anthropic has developed Claude Mythos Preview, their most capable frontier model to date, showing a striking leap over previous models like Claude Opus 4.6.
•Despite its advanced capabilities, Anthropic has decided *not* to make Mythos generally available due to significant safety concerns identified in its comprehensive System Card.
•The model scored high on various risk assessments, including chemical/biological, autonomy, and cybersecurity, prompting its limited deployment in a defensive cybersecurity program.
•Findings from Mythos's evaluations will directly inform the safety measures and release strategies for future Claude models, emphasizing Anthropic's commitment to responsible scaling.
•OpenAI has launched a new Safety Bug Bounty program dedicated to identifying AI abuse and safety risks.
•This program complements their existing Security Bug Bounty by accepting non-traditional vulnerabilities that pose real-world harm.
•Key focus areas include agentic risks (like prompt injection, data exfiltration), exposure of OpenAI proprietary information, and issues related to account and platform integrity.
•It's a call for the global security and safety research community to help secure rapidly evolving AI systems.
•OpenAI has released new prompt-based safety policies specifically designed to help developers build age-appropriate AI experiences for teens.
•These policies are built to integrate with `gpt-oss-safeguard`, OpenAI's open-weight safety model, simplifying the creation of safety classifiers.
•Developed with input from external experts like Common Sense Media, this initiative is a crucial step in operationalizing teen safety within AI applications.
