HigherEd AI Daily: April 10, 2026 – Claude Mythos Finds Thousands of Security Flaws, AI Vision Tools Raise Accessibility Questions, AlphaGenome Decodes the Dark Genome

Andrew Ng's newsletter today is one of the most important issues of the year for anyone working in research or education, and I have pulled the three stories that matter most for you.

Dr. Ali Green

Anthropic's Next Model Found Thousands of Undiscovered Security Vulnerabilities. Research Institutions Are Paying Attention.

Anthropic previewed Claude Mythos this week, a forthcoming model not yet available to the public. In a 244-page model card, Anthropic reported that Mythos autonomously discovered thousands of high-severity security vulnerabilities in major operating systems and browsers during testing, including a 27-year-old flaw in OpenBSD and a Linux kernel exploit. Both have since been patched. Anthropic assembled a 40-plus member consortium called Project Glasswing, including AWS, Apple, Google, and Microsoft, to harden open-source software before the model releases commercially.

Why this matters for your teaching

Research institutions that host data on standard operating systems or widely used software should pay attention to the timeline here. The good news is that Glasswing is actively patching what Mythos finds. This story is also a rich case study for courses on AI policy, cybersecurity, and responsible AI development.

Read more: The Batch by Andrew Ng (April 10 Issue)

Pulled from: The Batch by Andrew Ng

Blind Users Are Relying on AI to Assess Their Appearance. Researchers Say That Raises Hard Ethical Questions.

A BBC investigation highlighted this week that many blind and visually impaired people now use AI vision apps to evaluate how they look before meetings, classes, or dates. Apps including Be My Eyes and Microsoft Seeing AI sometimes offer subjective commentary that reflects cultural beauty standards embedded in training data. Psychologists at the University of Bristol warned that this can contribute to anxiety and depression, especially because blind users have no way to independently verify what the AI says about them.

Why this matters for your teaching

For faculty who teach courses on ethics, technology and society, or disability studies, this is a rich and immediate case study. It also raises a direct question for course designers: are the AI tools your students use accessible and equitable for students with disabilities?

Read more: The Batch by Andrew Ng (April 10 Issue)

Pulled from: The Batch by Andrew Ng

Google DeepMind's AlphaGenome Interprets 98 Percent of Human DNA That Science Previously Labeled Junk

Google DeepMind released AlphaGenome this week, an open-weights model that interprets non-coding DNA, the 98 percent of the human genome that does not directly produce proteins but regulates gene expression and cell behavior. The model accepts one million base pairs as input and outputs roughly 6,000 properties of the gene sequence. It outperformed nine previous models in 47 of 50 evaluations and is freely available for noncommercial research use.

Why this matters for your teaching

For faculty in biology, genetics, bioinformatics, and related fields, AlphaGenome makes a class of genomic analysis that previously required extensive wet lab work or specialized computing accessible through a free API. This is a meaningful research tool, not just a headline.

Read more: Google DeepMind AlphaGenome Release

Pulled from: The Batch by Andrew Ng

Try something new today

Prompt of the Day

"I teach a course on [AI ethics or disability studies or technology and society] and I want to design a case study discussion around AI vision tools used by blind and visually impaired people. The discussion should help students think critically about bias in training data, who gets to define beauty or normalcy, and what responsible AI design looks like for users who cannot verify AI output. Write five discussion questions and suggest one short reading or artifact students should engage with before class."

Tool of the Day

Be My Eyes

An AI-powered app that provides real-time visual assistance for blind and low-vision users. Useful for faculty exploring AI accessibility in research and teaching.

Free