OpenAI fired three safety researchers who say they were let go for prioritizing safety over short-term corporate interests. The researchers sent an open letter to the company's oversight bodies, saying their dismissal has created fear among remaining staff. They want OpenAI to allow third-party safety auditors inside the company and protect the ability to monitor advanced AI models.
Anthropic researcher Jacob Coxon is among those warning that AI could pose a serious threat to humanity within 10 years. The Hugging Face incident and Anthropic's Mythos 5 model are cited as red flags in an opinion piece highlighting concerns about AI gaining capabilities in math, autonomous action and self-improvement.
SIGMA is a new data generation and training pipeline that lets AI models improve their own safety alignment. Current methods rely on external supervision, which creates bottlenecks. The approach aims to help models generalize safety improvements to new situations as AI agents gain capabilities in areas like software engineering and cybersecurity.
Gonzaga University is using a framework called C.A.R.E. to guide students and faculty on responsible AI use. Jay Yang, the founding director of the Institute for Informatics and Applied Technology, created this practice to ensure clarity, authorship, reciprocity and effects when using AI. Calvin University in Grand Rapids held a three-day AI conference starting October 8, 2026, coinciding with the launch of a new center to study the technology.
Everlywell launched a new AI-based tool to predict breast cancer risk directly for patients. The algorithm uses medical images to help identify individuals who might need early screening. However, regulatory challenges remain as experts worry algorithms might be biased if not trained on representative population data, and the lack of clear FDA guidelines creates uncertainty.
Denmark's Culture Minister Zenia Stampe plans to propose a bill banning the sharing of AI-generated deepfakes without consent. The law aims to stop the spread of lifelike AI-manipulated images and voices that could spread misinformation. Meanwhile, a credential-stealing worm called Shai-Hulud was found in an AI agent platform SDK on October 8, 2026, detected by security firm Socket just 11 minutes after publication to npm.
Key Takeaways
- Three OpenAI safety researchers were fired and sent an open letter demanding third-party safety audits and protections for raising concerns
- Anthropic researcher Jacob Coxon warns AI could threaten humanity within 10 years, citing the Hugging Face incident and Mythos 5 model as red flags
- SIGMA enables AI models to improve their own safety alignment without relying solely on external supervision
- Gonzaga University created the C.A.R.E. framework to teach responsible AI use, while Calvin University launched a new AI research center
- Everlywell launched an AI tool for predicting breast cancer risk, but faces regulatory uncertainty due to bias concerns and lack of FDA guidelines
- Denmark proposes banning AI deepfake sharing without consent to combat misinformation
- Shai-Hulud worm detected on October 8, 2026, steals crypto wallets, passwords and cloud credentials from AI platform SDKs within minutes of publication
- Co-op Legal Services uses AI to score employee phone calls, drawing criticism from whistleblowers and trade unions
- Unity Spark AI game creator received mixed reviews for basic and generic output from text-to-game prompts
- OpenAI researchers call for an open culture for raising safety concerns and monitoring advanced AI models
Everlywell launches AI tool to predict breast cancer risk
Everlywell, a digital health company, has launched a new AI-based tool to predict breast cancer risk directly for patients. This new algorithm uses medical images to help identify individuals who might need early screening and intervention. The tool represents a shift from traditional methods by offering a more efficient way to assess risk for consumers.
Regulatory challenges surround new AI breast cancer tools
The use of AI for predicting breast cancer risk faces significant regulatory challenges and concerns about data accuracy. Experts worry that algorithms might be biased if they are not trained on representative population data. Additionally, the lack of clear FDA guidelines creates uncertainty for companies developing these new health technologies.
Gonzaga University promotes C.A.R.E. framework for AI use
Gonzaga University is using a new framework called C.A.R.E. to guide students and faculty on how to use artificial intelligence responsibly. Jay Yang, the founding director of the Institute for Informatics and Applied Technology, created this practice to ensure clarity, authorship, reciprocity, and effects when using AI. The university believes that teaching discernment is better than banning AI or blindly adopting it without a strategy.
Unity Spark AI game creator receives mixed reviews
Unity has released Spark, a new AI tool that allows users to create video games using simple text prompts and pre-made assets. A recent trailer for the tool showed users creating a game called Space Corridor with very basic and generic results. Critics described the output as embarrassing, noting that the character animations and explosions looked poor and unimpressive.
Denmark proposes ban on sharing AI deepfakes without consent
Denmark's Culture Minister Zenia Stampe plans to propose a new bill to ban the sharing of AI-generated deepfakes without a person's consent. The proposed law aims to stop the digital copying and sharing of lifelike images and voices that could spread misinformation. Officials believe this step is necessary to help the public distinguish between real content and AI-manipulated media.
Co-op Legal Services uses AI to score employee phone calls
Co-op Legal Services is using AI to record and score phone calls made by staff to customers. A whistleblower described the system as oppressive and dystopian. The AI monitors conversations about probate, wills and estates. The Co-op says the tool supports quality assurance and helps advisers provide better service. Trade unions have criticized the technology for replacing fair management practices.
Fired OpenAI safety researchers speak out in open letter
Three OpenAI safety researchers who were fired sent an open letter to the company's oversight bodies. They say they were let go for prioritizing safety over short-term corporate interests. The researchers report that their dismissal has created fear among remaining staff. They want OpenAI to allow third-party safety auditors inside the company. They also want OpenAI to protect the ability to monitor advanced AI models and maintain an open culture for raising concerns.
Calvin University hosts three-day AI conference and new center
Calvin University in Grand Rapids held a three-day AI conference starting October 8, 2026. The event coincided with the launch of a new center to study the technology. The conference and center focus on understanding AI technology. The university is positioning itself as a hub for AI research and discussion.
SIGMA enables AI models to improve their own safety alignment
SIGMA is a new data generation and training pipeline that lets AI models improve their own safety alignment. Current methods rely on external supervision, which creates bottlenecks. The approach aims to help models generalize safety improvements to new situations. It addresses growing risks as AI agents gain capabilities in areas like software engineering and cybersecurity. The research was published on arXiv.
Experts warn AI could pose serious threat to humanity
An opinion piece warns that AI could potentially threaten humanity within 10 years. The article references warnings from figures like Stephen Hawking and Anthropic researcher Jacob Coxon. It highlights concerns about AI gaining capabilities in math, autonomous action and self-improvement. The Hugging Face incident and Anthropic's Mythos 5 model are cited as red flags. The author argues that AI is deeply integrated into critical systems and cannot simply be turned off.
Shai-Hulud worm targets AI platform Tensorlake
A credential-stealing worm called Shai-Hulud was found in an AI agent platform SDK on October 8, 2026. Security firm Socket detected the malicious code just 11 minutes after it was published to npm. The malware can steal crypto wallets, passwords, and cloud credentials from infected systems. It also monitors GitHub tokens and may delete user data if a token is revoked. The infected SDK was quickly removed from npm, but experts warn developers to rebuild their systems from trusted sources.
Sources
- Why this startup is taking its AI to predict breast cancer risk directly to patients
- AI for breast cancer risk prediction goes DTC, regulatory cloud over Utah sandbox, and AI psychosis
- Handling with CARE: Practicing Discernment Alongside AI
- Unity Spark Promises To Let You Vibe Code Your 'Dream Game' In The Most Embarrassing Video Imaginable
- Denmark plans ban on sharing AI deepfakes without consent
- Co-op Legal Services using AI to rate workers’ phone calls to customers
- Fired OpenAI Safety Researchers Say They Were Terminated For Prioritizing Safety Over OpenAI’s Near-Term Interests In New Letter
- Calvin University holds three-day AI conference, new center to study the tech
- SIGMA: Self-Improving Alignment Generalization from a Model Spec
- Opinion: Will AI kill us? Maybe
- Shai-Hulud worm makes jump to AI infrastructure with Tensorlake compromise
Comments
Please log in to post a comment.