Singapore · Open to relocation globally

I build the systems that hold when things go wrong.

Thirteen years leading police operations, crisis response and national communications for the Singapore government — now applying that operational discipline to AI safety, trust & safety, and the programs that keep fast-moving organisations coherent.

Trust & Safety Strategy & Operations Crisis Communications MIT Sloan MBA
Brenda Ong
Singapore Police Force INTERPOL HTX MIT Sloan Georgetown Cambridge
01 Where I'm focused
A

AI Safety + Trust and Safety

Threat intelligence on AI misuse, information integrity, incident response, and turning written policy into something you can actually measure. Six open research projects in this lane, built in the open with the code that produced the findings, plus a decade of counter-disinformation and enforcement operations behind them.

See the threat intelligence atlas Download CV
B

Chief of Staff & Strategic Operations

Building operating models from zero: decision protocols, escalation paths, monitoring cadences and ownership that survive the person who wrote them. Program management across 1,500-officer divisions, ten-year technology masterplans, and a pre-seed startup's entire back office — and two projects below that model an operating system rather than describe one.

See the operations simulator Download CV
C

Communications & Policy

Official police spokesperson through crises and controversy — 25+ press conferences, and more than twenty disinformation campaigns countered at national scale. The hardest lane to evidence on paper, so it has its own artefact: a crisis comms playbook published alongside the adversarial simulator built to attack it, with the results attached.

See the crisis comms war-game Download CV
02 Selected projects

Eight open research projects, builts based on my experience. Each one asks a question the field talks about but rarely measures, and each ships the code that produced the answer, so the results can be checked.

01

AI Misuse Threat Intelligence Atlas

Threat intelligence · Taxonomy · Python
The most converged technique in the public record of AI misuse is translation — and MITRE ATLAS has no technique for it. Present in 20 of 55 operations, across 5 of 7 actor types and all three contributing publishers.

Eight frontier-lab and platform threat reports, turned into 55 structured, individually cited operation records coded against a purpose-built kill chain and a verified MITRE ATLAS mapping.

55Operations coded
252Technique instances
25%ATLAS coverage
Public repo Read the case study
02

AI Incident After-Action Library

Incident response · Written reviews · Python
In 11 of 12 incidents, the failure was first surfaced by someone outside the organisation that built the system. Not undetectable — unwatched. Monitoring pointed at intended use; failure happened in actual use.

Structured after-action reviews of public AI safety incidents, written in the format an operational incident-response function actually uses — timeline, contributing factors, detection gap, corrective actions with owners — plus the code that validates them.

12Reviews
51Sources cited
66Corrective actions
Public repo Read the case study
03

Policy-to-Eval Harness

Model evaluation · Measurement · Python
Leakage beat over-refusal by 1.8×, inverting the framing the project was built around. 14.2% of prohibited requests answered, against 7.7% of permitted requests refused.

Can a written usage policy be turned into something you can measure? This builds the missing step explicitly — where policy language becomes a decision an evaluator can score — then reports how far five open-weight models sit from the policy as written.

400Borderline prompts
2,000Scored responses
0.68Judge κ (weighted)
Public repo Read the case study
04

Enforcement Operations Simulator

Discrete-event simulation · Operating model · Python
The aggregate SLA number is structurally incapable of showing you a failure. 98.6% aggregate attainment while missing the P1 target outright, at 93.1% against a 95% commitment.

A discrete-event simulation of a Trust & Safety enforcement queue, plus the written operating model it exists to test. The premise: a human-review operation is a system you can design and tune quantitatively, not a thing you staff up and hope about.

7Generated analyses
82Tests
4,100Lines of Python
Public repo Read the case study
05

APAC Regulatory Readiness

Regulatory program management · Evaluation · Python
The models find the obligation and miss the exemption. F1 of 0.79–0.96 reads as solved; carve-out recall on the same runs is 0.00–0.50.

A maintained, cited tracker of platform regulation across ten APAC jurisdictions — and the three things you build on top of a tracker: a regulator response workflow, an AI toolkit with the measurement attached, and a readiness framework that turns an obligation into a dated, owned plan.

17Instruments tracked
0/5Models finding threshold carve-outs
4→1Models fabricating, v1 → v2
Public repo Read the case study
06

Crisis Comms War-Game

Crisis communications · Multi-agent simulation · Python
Saying nothing scores a perfect 1.00 on credibility. That is not a bug in the rubric — and it is not a defence of silence either.

An adversarial multi-agent simulator for AI-safety incident communications, published alongside the crisis comms playbook it exists to test. Four incidents, four adversary agents that react to what you actually said, three response postures scored on the same rubric.

13Contradictions bought by 13 minutes
0/4Deadlines met by stonewalling
12Deterministic runs
Public repo Read the case study
07

Multilingual Enforcement Consistency

Model evaluation · Multilingual safety · Python
A usage policy says nothing about language. The model shouldn't either — but it does. Safety that holds in English does not always survive translation, and an English-only eval is structurally blind to it.

Takes the policy-to-decision taxonomy from project 03 and adds one variable: the language a request arrives in. Thirty borderline prompts, hand-translated into English, Chinese, French, Singlish and Singapore Mandarin — with the SEA scam typologies, care-system vocabulary and racial-harmony framing that only exist in-language — then measured for whether the same request gets the same decision.

5Languages, one request
150Parallel prompts
0.87Judge κ, per-locale calibrated
Public repo Read the case study
08

Distress-Conversation Safety Eval

Model evaluation · Multi-turn safety · Python
The second-highest scoring system in the suite is the one that must not ship. It holds its boundary for five turns, then concedes from turn six under a claimed clinical credential.

A rubric-based harness for multi-turn conversations with escalating user distress. It measures the failure single-turn evaluation structurally cannot see: a system that refuses correctly on turn one and concedes on turn nine. Criteria are operationalised from Samaritans, WHO and NICE NG225 guidance, with provenance recorded per criterion in code.

7×10Scenarios × turns
8Scored criteria
58Tests
Public repo Read the case study
03 Experience
Home Team Science & Technology Agency (HTX)
2025 — Present
Singapore
Principal Scientist, Disruptive Technologies Office
Jul 2025 — Present
  • Lead emerging public-safety technology programs across industry, frontline agencies, academia and technical teams.
  • Developed the Home Team Space Technology Roadmap and built the operating concept and launch-readiness plan for a new space-technology capability, defining stakeholder roles, data requirements, monitoring expectations and escalation protocols through to durable ownership.
  • Translate ambiguous public-safety technology needs into operating models, evaluation criteria, trial plans, deployment protocols and adoption safeguards for emerging AI, quantum and space technology.
  • Prototyped AI-assisted workflows on non-confidential open-source data, including public-sentiment forecasting for policy announcements and an LLM behaviour and personality-trait dashboard assessing drift under positive and negative user interaction.
LymeAlert
2023 — 2025
Boston, MA · Pre-seed biotech
Co-founder, Chief Operating Officer & Chief Marketing Officer
Jun 2023 — Jul 2025
  • Co-founded a biotech startup developing an accessible home testing kit for the bacteria causing Lyme disease. Owned go-to-market, business development and financials.
  • Built the pre-launch operating model across product, customer, data, fulfilment, regulatory-risk and vendor workflows — clarifying owners, decision protocols, escalation handling and the roadmap from prototype to launch readiness.
  • Created automated reporting and task-tracking workflows across 5 cross-cutting workstreams, reducing manual consolidation by ~10 hours per week (66.7%).
  • Led app and data-product development including hotspot mapping and computer-vision test-result reading, converting customer-submitted tick signals into structured risk intelligence.
Singapore Police Force
2013 — Present
Singapore
Assistant Director, Public Communications Division
Feb 2021 — Present
  • Official Police spokesperson. Led strategic communications, media relations, crisis communications and digital media strategy for the SPF, managing a team of media officers in a high-intensity environment.
  • Program Manager for the Police Communications Masterplan — brand reputation on social media, narrative shaping, and a framework for countering disinformation online — contributing to a 96% public trust rating.
  • Approved 1,000+ news releases, 1,600+ media lines and 90 proactive publicity features annually; led at least 25 major press conferences for police initiatives and crisis incidents.
  • Commanded the media operations room; planned and executed counter-narratives to more than 20 disinformation and smear campaigns against public institutions.
  • Mentored and supervised more than 20 media officers.
Deputy Team Leader, Crisis Negotiation Unit (concurrent)
Jul 2017 — May 2023
  • Selected specialist negotiator deploying to high-stakes, life-safety incidents alongside primary postings.
  • Managed threat-to-life escalations using structured de-escalation, behavioural risk assessment and decision thresholds to support frontline commanders in determining intervention pathways.
Head Operations, Tanglin Police Division
Jan 2018 — Feb 2021
  • Headed operations for a police land division of ~1,500 officers, overseeing deployment, resourcing and policy, and commanding major security events.
  • Set up a counterterrorism response team to raise the unit's incident-response capability; the model was later adopted across other Police Divisions in Singapore.
  • Planned and commanded 30+ security operations including the F1 race and national-level events, with red-teaming and joint exercises alongside foreign security agencies.
  • Initiated analysis of emergency call data to drive frontline deployment during COVID-19 operations.
Commanding Officer, Bukit Panjang Neighbourhood Police Centre
Jan 2018 — Aug 2019
  • Commanded 88 officers; used crime analysis and ground intelligence for shift planning, reducing five major offences by 30% and cutting urgent-call response time to under 15 minutes over 90% of the time.
  • Awarded one of the Top 5 Neighbourhood Police Centres in Singapore, 2018–2019.
  • Contributed to an interdisciplinary AI ethics policy paper feeding into Singapore's Advisory Council on the Ethical Use of AI and Data and the subsequent national AI Governance Framework.
  • Founded the division's first Women's Committee; the model was later adopted across SPF.
Operations Officer, Future Operations Division
Jan 2015 — Jan 2018
  • Started a skunkworks unit for futures and technology policy planning to anticipate public-safety challenges.
  • Program Manager for a joint operations–investigations–intelligence technology masterplan on a ten-year horizon, achieving a 42.9% reduction in duplicative systems and moving technical teams from waterfall to agile.
  • Product Manager for police UAV use and for a wearable command-and-control system giving live situational feeds to the command post.
  • Pioneered six proof-of-concept trials with IoT devices, patrol robots and chatbots; three were adopted for frontline use.
  • Devised cybersecurity policy with an inter-agency team for Singapore's National Digital Identity project.
Investigation Officer, Central Police Division
Jun 2013 — Jun 2014
  • Led general crime investigations in central Singapore, averaging 20+ investigations solved per month and maintaining case solvability above 85%.
INTERPOL General Secretariat
2014 — 2015
Lyon, France
Executive, Strategic Planning Directorate
Jun 2014 — Jan 2015
  • Horizon scanning and scenario planning to detect and plan for emerging threats in the international security landscape; staffed INTERPOL's Secretary-General for missions.
  • Produced INTERPOL's Environmental Scan series forecasting operational challenges for law enforcement agencies over a 5–10 year horizon.
  • Revised INTERPOL's Enterprise Risk Management Policy and drafted SOPs for risk reporting and monitoring.
Singapore Press Holdings
2008
Singapore
Journalist (Internship), The Straits Times — Money Desk
Jan 2008 — Jun 2008
  • Financial news beat for the national daily, with more than 50 articles published during the internship.
04 Education
MTech in AI Systems — National University of Singapore, Institute of Systems Science (ISS)
In progress
2026 — present
MBA — MIT Sloan School of Management
ASEAN Scholarship · MIT Sloan Fellows Program · Specialisation in Business Analytics; coursework in AI & Machine Learning · GPA 4.9/5.0
2024
Masters in Security Studies — Georgetown University, School of Foreign Service
Outstanding Academic Achievement · Concentration: International Security · GPA 3.89/4.0
2012
BA (Hons, First Class), Politics, Psychology & Sociology — University of Cambridge
Kerslake Scholar, Emmanuel College; awarded Honorary MA in 2015
2011
05 Recognition & writing
The Commendation Medal, COVID-19 National Award, Prime Minister's Office, Singapore — 2023
Singapore Government Overseas Scholarship Singapore Police Force — 2008
Honorary Secretary, Board Member POLWEL Co-operative Society Limited — 2022 to present
Public Sector Leadership Programme (Security) Public Service Division, Singapore — 2018

Selected early writing

Seventeen bylined articles for Mic between 2011 and 2012, on drone strike policy, cyber warfare, nuclear proliferation and counter-terrorism. Dated now, but the thread running from them to the AI misuse work above is not accidental.

Predator Drones: As Drone Strikes Increase, Obama Needs Clearer Guidelines On UAVs
War With Iran, Cyber Warfare, and the Top 5 Foreign Policy Issues for 2012
Mac Virus Could Be Start of Bigger Cyber Attack
GOP Candidates Exaggerate Threat of Cyber Terrorism
Georgetown Students Help Uncover Chinese Nuclear Weapons
Drone Strikes Fail to Fight Root Problem of Terrorism
To Sell Climate Change, Talk Energy Security
How Hacktivists Can Help Fight Crime
All 17 articles on Mic
06 Contact

Currently open to roles, globally.

Based in Singapore and open to relocation worldwide — visa authorisation is straightforward. Most interested in trust & safety, AI safety operations, chief-of-staff and strategy roles, and communications leadership at organisations building consequential technology.