About Alt-TAB
The most comprehensive free, publicly accessible pre-deployment ethics and Safety by Design assessment for AI and emerging technology. Built in Australia. Launched at the United Nations. Used around the world.
Why Alt-TAB exists
Preventable digital harm should never be the cost of innovation.
Most harm caused by technology was preventable. It just wasn't looked for. The failures that make headlines rarely come from bad intentions. They come from good people who never stopped to ask the right questions before they shipped something.
The build cycle has no natural moment where someone stops and asks who this could hurt. Sprints move fast. Deadlines compress. Ethics conversations get deferred until after launch, when they're hardest and most expensive to act on. Alt-TAB creates that moment. Before deployment, when something can still be changed.
It's not a compliance checklist. It's named after the keyboard shortcut that switches windows: a designed pause, an intentional interruption. The questions matter. But the pause is the intervention.
Three things every assessment addresses
1
Cybersecurity and systems integrity
Data flows, threat vectors, access controls, and potential exploitation pathways. Digital harm is not only a social issue. It's a systems vulnerability, and it gets checked in every assessment.
2
Human rights and equity safeguards
Gender equality, child protection, disability access, Indigenous data sovereignty, and inclusion. Who does this affect? Who gets left out? Who is carrying a risk they don't know about?
3
Technology-facilitated gender-based violence prevention
AI tools, data systems, and digital platforms can be weaponised for coercion, exploitation, and abuse. This is screened in every single assessment regardless of what kind of product is being built, because the risk is almost always invisible to the builder.
The founder
Sarah Barnbrook
Founder & CEO, Away from Keyboard Inc., Melbourne, Australia
I built Alt-TAB because I spent years working where technology, family violence, and child safety intersect. The same pattern kept showing up. By the time harm was visible, it was already too late to prevent it. The systems meant to protect people had been designed without ever asking who they might hurt.
That work took me from community practice in Melbourne to the United Nations Commission on the Status of Women in New York, to the IEEE where I serve as Co-Vice Chair of the Industry Connections Activity on AI Used in Evaluating Family Violence, and to Geneva where I am speaking at WSIS and the AI for Good Global Summit in July 2026.
I hold a Turing College AI Ethics certification developed with University College Dublin and Sciences Po, EU-funded and aligned with the EU AI Act. Alt-TAB is what that work looks like in practice: free, open-source, and grounded in 30 real governance frameworks, because thinking carefully about technology shouldn't require a budget or a specialist team.
🇺🇳 UN CSW70, New York 2026
🎙️ WSIS Geneva, July 2026
🎙️ AI for Good Global Summit, Geneva 2026
⚡ IEEE IC25-008 Co-Vice Chair
🎓 Turing College AI Ethics, UCD and Sciences Po
🏛️ Accredited UN Delegate, CSW
WSIS Geneva, July 2026
"Inclusion Without Safety Is Not Empowerment: Automation Is Scaling Harm Faster Than We Can Respond"
"
Innovation without inclusion isn't progress. It risks building systems that overlook the very people they are meant to serve.
— Sarah Barnbrook, NGO CSW70 Parallel Event
Reimagining the Possible: Women, AI and a Fairer Future
CCUN, New York City, March 2026
Recognition & reach
Alt-TAB launched on 12 March 2026 at the United Nations Commission on the Status of Women in New York City. The workshop was called Signals of Safety: When Technology Listens to Women, exploring how ethical AI and digital tools can detect coercive control, bias, and online harm when Safety by Design principles are built in from the start. Read the launch story →
In July 2026, Alt-TAB and its approach to upstream harm prevention will be presented at two major international forums in Geneva: the World Summit on the Information Society (WSIS), and the AI for Good Global Summit. The session at WSIS is titled "Inclusion Without Safety Is Not Empowerment: Automation Is Scaling Harm Faster Than We Can Respond."
Since launch, Alt-TAB has been used by startup founders, NGOs, educators, government teams, and researchers across Australia and internationally to think through what they might be missing before they deploy.
Child safeguarding and upstream harm prevention
Child safety is Alt-TAB's highest-risk category. Unlike other risk domains, it's screened in every single assessment, regardless of whether the builder mentions children. This is a deliberate design decision grounded in how harm actually occurs: children encounter technology designed for adults. A fitness app, a community forum, a local business booking system, a mental health tool. None of these are "children's products," and almost none of their builders consider child safety when designing them. Alt-TAB does.
In the dataset of 146 assessments conducted since launch, child safety gaps (missing age verification, features enabling adult-child anonymous contact, platforms with community features that create grooming pathways) represent the most serious harm potential of any risk category identified. The mandatory screening means these gaps surface even when the builder has not considered children at all.
The question Alt-TAB asks is not "is this a children's product?" but "could a child be harmed by this?" That reframing is the intervention.
Upstream harm prevention
Upstream prevention means stepping in before harm happens, not cleaning up after it. In child protection, this principle is well established in social work and public health. It's rarely been applied to technology design. Until now. Alt-TAB creates the moment where a builder stops and asks who this could hurt, before the product ships, when something can still be changed.
Technology-facilitated child exploitation almost never involves malicious code. It exploits legitimate features: messaging systems, anonymous profiles, community forums, location data, content recommendation algorithms. These features are built without malicious intent and harm children anyway. Alt-TAB uses ACCCE research on how these harm pathways actually operate to ask the questions that surface vulnerabilities before a child is harmed, not after.
Child safety frameworks applied in every relevant assessment
UN Convention on the Rights of the Child
The foundational international treaty on children's rights, ratified by Australia. Articles 3 (best interests of the child), 16 (privacy), and 17 (access to safe information) applied directly to technology design decisions.
ACCCE Research on Child Exploitation
The Australian Centre to Counter Child Exploitation produces evidence-based research on how digital platforms are exploited for grooming and child sexual abuse material. This research directly informs the follow-up questions Alt-TAB generates for products with child exposure risk.
UNICEF Guidance on AI and Children 3.0
UNICEF's 2025 framework for responsible AI when children are involved, covering data collection, consent, age-appropriate design, and AI systems that interact with or affect young people.
UK Age Appropriate Design Code
The most detailed practical standard for child-safe product design in existence. 15 design standards including privacy by default, no nudge techniques, and data minimisation. Cited as a design benchmark for Australian products and as a legal obligation where UK child users are involved.
How it works
1
You tell us what kind of decision you're making
Before any questions, you pick the type of technology decision you're facing: building something from scratch, adopting an existing tool, changing a digital process, or reviewing something already running. This changes everything: the questions you get, the language used, and what the AI focuses on. A community organisation moving client records gets a completely different experience than a startup building an AI product, even if they both flag the same risk areas.
2
Stage 1: nine questions, four layers of analysis
You answer nine domain questions and describe your product or process in plain language. That description is the most important thing you submit. It's what the AI checks your answers against.
The Stage 1 analysis works in four layers:
Consistency checking. If you say you have safeguards but your description doesn't mention any, the AI flags that as confident ignorance, one of the four risk patterns the tool looks for. You can't game it by ticking boxes, because the description is the ground truth.
Mandatory screening. Two areas are checked in every single assessment regardless of what you answer: technology-facilitated gender-based violence and cybersecurity. These risks are almost always invisible to builders. Not because they're careless, but because the harm pathways aren't obvious. A fitness app with location sharing is a technology-facilitated gender-based violence risk. A community calendar with user accounts is a cybersecurity exposure. Child safety is also screened in every assessment, because children encounter technology designed for adults and most builders never consider this.
Jurisdiction calibration. The tool reads your description for context: where you're based, who your users are, whether First Nations communities or children are involved. Australian products get Australian law. Products with EU users get GDPR flagged. Products affecting Aboriginal and Torres Strait Islander communities get AIATSIS and CARE Principles guidance.
Tailored follow-up questions. For each risk area flagged, the AI writes three to five follow-up questions specific to your product. These aren't drawn from a question bank. They're written fresh for each assessment based on what your description revealed.
3
Stage 2: your full readiness report
You answer the follow-up questions and tell us your deployment timeline. A second AI analysis takes everything: your original answers, your description, your follow-up responses, and your timeline. It produces a structured report with:
A readiness score out of 100 with a plain-language explanation of how it was calculated. Not an arbitrary number, but one derived from the specific risks found, safeguards identified, accountability structure, and how soon you're planning to launch.
Blindspot analysis where each gap is put into one of four categories: confident ignorance, missing context, false safety, or second-order risks. Each one includes the evidence from your submission, a question to ask yourself, and the framework that applies.
Specific recommendations clearly split between legal obligations and ethical best practice, so you can see what the law requires versus what's simply the right thing to do. Every recommendation includes a concrete first step in plain language.
A 30-day action plan split into this week and this month, timed to your deployment.
Questions to take to your team, written for the person running the assessment to hand to whoever can actually act on the findings.
4
You get a PDF and nothing gets stored
The full report downloads as a PDF, including the action plan, team questions, and all. Legal obligations are visually separated from ethical recommendations. Every framework cited links to a plain-language explainer. And nothing you submitted is kept: no product descriptions, no answers, no report content. Only anonymised stats are retained, and only with your consent. Alt-TAB was built to practise what it assesses.
What makes this different from other tools
35 frameworks at once
Most ethics tools cite one or two frameworks. Alt-TAB applies 30 simultaneously, from the UNCRC and the Disability Discrimination Act 1992 to GDPR and the eSafety First Nations Family Safety resource. The relevant ones are selected based on what your description reveals.
You can't just tick boxes
The consistency detection layer checks your description against your answers. If they don't match, the gap gets flagged. It's the feature that matters most, because most self-reported assessments can be passed by someone who's confident but wrong.
Four risk types, not a list of problems
Gaps are put into one of four categories: confident ignorance, missing context, false safety, or second-order risks. This taxonomy was developed through analysis of 146 real assessments and refined through the complete redesign of the tool. It tells you not just what's wrong but what kind of wrong it is.
The law that applies to you specifically
Australian products get Australian law. EU users trigger GDPR. First Nations community impact triggers AIATSIS and CARE Principles. The recommendations aren't generic. They're built from what your description reveals about your context and reach.
Built for the person actually doing it
A community group migrating a contact list has different risk exposure than a startup building an AI product. The tool knows the difference and adjusts, not just cosmetically, but in the questions asked, the frameworks applied, and the language used.
Actually private
What you describe is never stored. The analysis happens in real time and the content is discarded. This isn't a policy. It's how it's built. We couldn't access your product description after the fact even if we wanted to.
How it makes a difference
Alt-TAB makes impact in three ways.
Better decisions before launch
By surfacing threat indicators, misuse pathways, exploitation risks, coercive control vectors, and unsafe data practices before deployment, the tool helps organisations identify gaps between what they think their risk is and what it actually is.
Access to ethical governance for everyone
As a free tool, it enables small businesses, startups, educators, and community organisations to apply structured risk logic without needing dedicated legal, cybersecurity, or compliance teams. Ethics governance shouldn't be a service only well-resourced organisations can afford.
Cultural change over time
By normalising upstream ethical reflection as part of innovation, Alt-TAB helps reframe technology development from "move fast" to "build responsibly." When organisations learn to identify harm pathways before they exist, innovation becomes more equitable, more resilient, and more trustworthy.
Alt-TAB is not anti-innovation. It's pro-responsibility. It demonstrates that you can move fast and still build safely, when safety is embedded from the start rather than bolted on after the fact.
Alt-TAB checks your situation against 35 Australian and international governance frameworks, chosen based on your jurisdiction and what comes up in your assessment. Two risk areas, technology-facilitated gender-based violence and cybersecurity, are checked in every single assessment regardless of your answers, because these risks are almost always invisible to the people building the thing.
MARAM: Multi Agency Risk Assessment and Management Framework
Family Violence Protection Act 2008 (Victoria)
Our Watch: Change the Story Framework
National Plan to End Violence Against Women and Children 2022-2032
Safe and Equal MARAM Practice Guides
UN Convention on the Rights of the Child
UNICEF Guidance on AI & Children 3.0
UK Age Appropriate Design Code
ACCCE Child Exploitation Research
UNESCO Recommendation on the Ethics of AI
OECD AI Principles
NIST AI Risk Management Framework
UN Guiding Principles on Business & Human Rights
Global Digital Compact (UN)
Privacy Act 1988 & Australian Privacy Principles
Online Safety Act 2021 (Australia)
eSafety Commissioner Safety by Design
eSafety Technology-Facilitated Gender-Based Violence Industry Guide
eSafety Phase 2 Industry Codes
eSafety Self-Harm Material Guidance (2026)
eSafety Women in the Spotlight
eSafety First Nations Family Safety
ASD Essential Eight
NAIC Voluntary AI Safety Standard
NAIC Practical Guides and Learning Hub
National Framework for Assurance of AI in Government
eSafety Tech Trends and Challenges
UN Convention on the Rights of Persons with Disabilities
Disability Discrimination Act 1992 (Australia)
WCAG 2.2 Accessibility Guidelines
Australian Digital Accessibility Toolkit
CARE Principles for Indigenous Data Governance
AIATSIS Code of Ethics
Design Justice Network Principles
Childlight / University of Edinburgh Research
From prototype to what it is today
The foundation of Alt-TAB was laid through the Victorian Summer of Cyber program, a research initiative run by the Australian Women in Security Network. Between January and April 2026, two student researchers, Camille Ang and Hashini Thanushika, worked with Away from Keyboard under supervision from Dr Muna Al-Hawawreh to review governance frameworks, conduct early methodology development, build a first digital prototype, and run alpha and beta testing with real users.
That work was valuable. It confirmed what the evidence already suggested: governance frameworks exist, but most organisations making technology decisions can't access or interpret them. The prototype demonstrated real demand for a practical, accessible tool. User feedback was clear about what was missing: the outputs were the same regardless of what you put in. Every scenario got essentially the same generic response. That wasn't good enough.
The current version of Alt-TAB is a complete redesign built by Sarah Barnbrook after the program concluded. The prototype was a starting point. What exists today is something fundamentally different.
The two-stage AI analysis engine, the consistency detection layer that checks your answers against your description, the four risk pattern taxonomy, the 35 frameworks applied simultaneously, the jurisdiction-aware recommendations, the mandatory technology-facilitated gender-based violence and child safety screening in every assessment regardless of what you say, the tailored follow-up questions written fresh for each submission, and the legal versus ethical separation in the report, none of that existed in the prototype. It was designed, built, and deployed by Sarah as a complete rebuild.
Camille and Hashini laid important groundwork. The current tool represents what became possible when that groundwork was rebuilt from the inside out.
Original prototype supervised by Dr Muna Al-Hawawreh. Current tool designed and built by Sarah Barnbrook, Away from Keyboard Inc. Full research report available on request.