Tendril
Loading…
New · guided experience
A curated walkthrough of the library — ordered lessons, a 15-question quiz between each, and 3 next-steps so you stay in flow. Earn XP, badges, and a streak as you go.
Library · 6440 lessons · Safety & Ethics view
Bias, privacy, copyright, and staying in control. Pick a tool lane below to drill into concrete workflows, then use the browser to refine by age, situation, or skill level.
Drill down
Start with a real app or workflow. Each lane filters the library to practical lessons, not just broad theory.
548 lessons in safety & ethics
Lessons handpicked for the Safety & Ethics shelf.
LLMs inherit the skews of their training data and RLHF feedback. Auditing for bias isn't a one-time test — it's an ongoing practice that belongs in every deployment.
Training data copyright is actively litigated. While courts work it out, deployers face practical decisions about outputs that copy protected material.
AI-generated media has crossed the perceptual threshold where humans cannot reliably detect it. Detection tools help — but are in an arms race with generation.
Prompt injection is the SQL injection of the AI era — and it's already being exploited in production systems. Defending against it requires multiple layers, not a single fix.
Fresh Safety & Ethics lessons added to the library.
Where bias comes from, what mitigation can and cannot do, and what to watch for.
What works locally now, what does not, and why it matters.
Why models refuse what they refuse, and how that shapes their behavior.
AI helps creators draft emergency handover documents so the channel doesn't disappear if they're suddenly unavailable.
Browse everything
Subject tracks
Tap a tile to filter the library — or pick “Surprise me” below for a randomized starter set.
LLMs inherit the skews of their training data and RLHF feedback. Auditing for bias isn't a one-time test — it's an ongoing practice that belongs in every deployment.
Training data copyright is actively litigated. While courts work it out, deployers face practical decisions about outputs that copy protected material.
AI-generated media has crossed the perceptual threshold where humans cannot reliably detect it. Detection tools help — but are in an arms race with generation.
Prompt injection is the SQL injection of the AI era — and it's already being exploited in production systems. Defending against it requires multiple layers, not a single fix.
Jailbreaks are how deployed AI systems fail publicly. Red-teaming is how you find those failures in private first — and it's a discipline, not a one-day exercise.
AI deployment in workplaces raises consent questions that legal minimums don't fully address. Employers who lead on transparency gain trust; those who don't face backlash.
Model cards and transparency reports are how AI providers document what their systems can and can't do. Knowing how to read them — and what's missing — is a core deployer skill.
The EU AI Act is the world's first comprehensive AI regulation, and its effects reach well beyond Europe. Here's what deployers worldwide need to understand right now.
Publishing AI research or releasing models creates benefits and risks simultaneously. The norms for when to disclose, delay, or withhold are evolving — deployers need a framework.
Training large models makes headlines, but inference runs constantly. The environmental cost of AI at scale is a design constraint as much as a compliance question.
Bias audits run once at deployment miss everything that emerges in production — distribution shift, edge-case interactions, fairness drift. A real audit pipeline runs continuously and surfaces issues to humans for evaluation.
Effective AI red-teaming goes beyond clever prompts. The exercises that surface real risk include socio-technical scenarios, integration-point attacks, and post-deployment misuse patterns.
Jailbreak techniques evolve weekly. A jailbreak test suite that doesn't update is fossilized within months. Here's how to design a testing methodology that learns from the public attack landscape.
Poisoned training data — whether from compromised supply chains or insider attacks — can introduce backdoors that survive evaluation. Detection requires provenance tracking, statistical anomaly detection, and behavioral evaluation against trigger patterns.
AI system incidents — bias failures, safety failures, model behavior changes — require a different incident response than traditional outages. Here's the runbook your team needs before the next incident hits.
Training and deploying AI across borders triggers a maze of data protection regimes. Compliance isn't optional — and the rules are tightening, not loosening.
Most AI vendor security questionnaires miss the AI-specific risks. Here's the question set that surfaces vendors with real safety practice from those with marketing veneer.
Modern AI deployments stack 5-10 vendor models, libraries, and services. When something goes wrong, you need to know exactly what's running where. Here's how to maintain real attestation.
Public AI benchmarks (MMLU, HumanEval, etc.) tell you general capability. Private evals on your data tell you actual production fit. The smart teams maintain both.
Some AI failures harm users and warrant public disclosure. Knowing when (and how) to disclose is its own discipline — far beyond the standard breach-notification playbook.
An AI classifier with 95% overall accuracy can have 70% accuracy for one demographic and 99% for another. Subgroup fairness evaluation is what catches this.
Watermarking AI-generated content is a partial solution to provenance. The current state is messy: standards are emerging, adoption is fragmented, removal is possible.
AI productivity-monitoring tools have exploded. The research shows they often hurt the productivity they're meant to measure — while damaging trust permanently.
Your vendor's AI incident becomes your incident. Knowing your obligations to your own users — disclosure, remediation, credit — matters before the vendor's incident hits.
AI deployments with child users hit COPPA, state child-protection laws, and an evolving safety landscape. The compliance bar is substantially higher than adult-AI deployment.
AI helps make medical decisions every day. When something goes wrong, who's responsible? The legal answers are still forming — but practical risk allocation patterns are emerging.
Boards are asking about AI risk. Most reports they get are technical noise. Here's what board members actually need to oversee AI well.
Government AI procurement carries elevated transparency, fairness, and accountability requirements. The procurement process itself encodes the public interest.
Recommendation AI optimized for engagement can promote harmful content. Designing systems that resist this requires deliberate trade-offs.
AI in elder care can reduce isolation and improve safety — or strip dignity and create new harms. The design choices matter enormously.
News organizations using AI for production, personalization, and translation face trust trade-offs. Disclosure and editorial judgment remain primary.
AI in tenant screening, mortgage decisioning, and rental pricing faces strict Fair Housing Act compliance. Disparate-impact tests are the standard.
Federal and state laws now require AI disclosure in political advertising. Compliance evolves rapidly — and enforcement is ramping up.
Shadow AI happens when employees deploy AI without IT/security knowledge. Inventorying is the first step to managing it.
When AI recommendations affect people's lives (jobs, loans, housing, healthcare), explanations are required — by law and by trust.
Vendor AI incidents become your incidents. Researching vendor incident history before signing protects against repeat exposure.