About
I’m the founder of GT Systems, an Australian deep-tech company pioneering the development…
Articles by Rhett
Activity
2K followers
-
Rhett Sampson shared thisDay two. Another one. Once more with feeling: this is another OAuth issue and OAuth breaks under AI. SPAN uses defence grade cryptographic security.Rhett Sampson shared thisWARNING: Lovable is experiencing a mass data leak. If you built a project before Nov 2025, your source code, database credentials & AI chat history can be viewed by any free account today. Here's how to triage ASAP 👇 Hell of way to start a week lol Let's just treat this as a fun "part two" to yesterday's fiasco... (And don't get me started on the vibe coding tax). Anyway... This is the result of a Broken Object Level Authorization (BOLA) flaw in Lovable's API. Firebase auth tokens are checked, but ownership is never verified. Any free account can query any project. Affected endpoints: /projects/{id}/*, /git/files, /git/file, /documents. All return 200 OK for pre-November 2025 projects. A researcher (weezerOSINT) created a free account and pulled the full source tree of an admin panel built for Connected Women in AI a real Danish nonprofit. 3,703 edits this year. Last touched 10 days ago. The source contained hardcoded SUPABASE_URL, SUPABASE_PUBLISHABLE_KEY, and SUPABASE_SERVICE_ROLE_KEY. The service role key bypasses every Supabase security policy. The researcher pulled real names, companies, LI profiles of speakers from Accenture Denmark and Copenhagen Business School. This was reported to Lovable on March 3 via HackerOne (report #3583821). Lovable shipped ownership checks for NEW projects and left every existing project exposed. A project created in April 2026 returns 403 Forbidden. 48 days later, the HackerOne report is still open. Per the researcher, affected accounts include employees at Nvidia, Microsoft, Uber, and Spotify. And every AI conversation (schemas, error logs, credentials, and business logic you, etc is stored and reachable through the same flaw). This is Lovable's second major security incident in 12 months. What to do: 1 / Understand what "private" actually means on Lovable Legacy projects are readable by any free account via the public API. Lovable is calling this "intentional behavior." Their recommended fix is set your project to private. That is a paid-tier feature. 2 / Rotate everything that ever touched a Lovable project built before Nov 2025 Rotate Supabase service role keys FIRST. These bypass RLS entirely Rotate every database credential (connection strings, anon keys, publishable keys) Rotate every API key you pasted into a Lovable AI chat for debugging Rotate every third-party token in your source (Stripe, SendGrid, OpenAI, auth providers) Assume customer data in your Lovable-provisioned Supabase has already been enumerated 3 / Audit your actual exposure Open your project URL in incognito. If you can load the code / chat history, so can a stranger. Check Supabase logs for anomalous read queries on user, billing, or admin tables over the last 48 days Check your Lovable project's request logs for traffic to /projects/*, /git/files, /git/file, and /documents endpoints from unfamiliar sessions If you ever pasted a credential into a Lovable chat, treat it as compromised.
-
Rhett Sampson shared thisThis is what happens when you use AI agent frameworks that depend on OAuth. SPAN uses cryptographically anchored ID, permissions, policy, authorisation and domain security. No OAuth.Rhett Sampson shared thisWARNING: Vercel got hacked. If you've OAuth'd AI tools into Google, assume your source code, API keys, and customer data is being sold on the dark web today. Here's how to triage ASAP👇 The attack vector was a compromised third-party AI tool (Context . ai) with a Google Workspace OAuth app. One OAuth grant led to one Vercel employee account, which led to source code, NPM tokens, GitHub tokens, and production secrets for a platform hosting a meaningful percentage of the modern web. Here's what to do right now: 1 / Check if you're exposed (this takes maybe 60 seconds) 1. Go to admin . google . com 2. Security, then Access and data control, then API controls, then App access control, then Manage Third-Party App Access 3. Search for this OAuth Client ID: 110671459871-30f1spbu0hptbs60cb4vsmv79i7bbvqj {dot} apps {dot} googleusercontent {dot} com 4. If found, revoke and block immediately 2 / Vercel cleanup (assume compromise) Anything NOT marked as a "sensitive environment variable" should be treated as leaked. 1. Rotate every API key, DB credential, and token stored in Vercel env vars 2. Review your org's activity log for unusual deploys, invites, or token grants 3. Audit org members and remove anyone who shouldn't be there 4. Disconnect your GitHub integration, then reconnect it (invalidates stale tokens) 3 / Supply chain ShinyHunters claims to hold Vercel's NPM and GitHub tokens. If true, a poisoned release of next, @vercel/*, or any Vercel-published package is on the table. 1. Pin every Vercel-published dependency to an explicit, verified version 2. Rotate any credential that ever touched a Vercel build pipeline 3. Monitor build and deploy logs for anomalous steps Third-party OAuth apps that request scopes beyond the basics (name, email, profile pic) are a major new attack vector. Ask your Google admin to restrict "unconfigured third-party apps" to basic-info-only scopes. This won't be the last hack.
-
Rhett Sampson shared thisThis is a massive signal. The AI agent economy is real. Now we just need the AI agent fabric. #SPAN Salesforce launches Headless 360 to turn its entire platform into infrastructure for AI agents | VentureBeat https://lnkd.in/gvEnUQHBSalesforce launches Headless 360 to turn its entire platform into infrastructure for AI agentsSalesforce launches Headless 360 to turn its entire platform into infrastructure for AI agents
-
Rhett Sampson shared thisMark is spot on here. Then theres coordination beyond orchestration. And then theres coordination as infrastructure. That's the bottom line.Rhett Sampson shared thisEveryone's betting on AI. Most are betting on the wrong layer. The week's venture deals tell a clear story about where capital thinks the value sits right now: Layer 1 (Foundation models). OpenAI has raised $120B in total. Effectively infrastructure. The bet here is enormous, but the margin story is already getting complicated. Layer 2 (Vertical agents) - Harvey (legal) at $11B. Basis (accounting) just hit a unicorn valuation of $1.15B. These are agentic systems replacing knowledge-work workflows, not augmenting them. Strong early traction where buyers are motivated (BigLaw, enterprise finance). Layer 3 (Orchestration) - The quietest layer, and probably the most interesting. Who manages multi-agent workflows? Who audits them, governs them, and ensures they don't hallucinate at scale? Gartner forecasts 40% of enterprise applications will embed task-specific agents by the end of 2026, up from under 5% last year (https://lnkd.in/gdgTYw8e). If that's even half right, the orchestration layer becomes essential infrastructure fast. The foundation model race is largely decided. The vertical agent race is early but crowded. Orchestration is where the next defensible position gets built.Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025
-
Rhett Sampson shared thisThis is EXACTLY what SPAN does. It is the semantic layer Maria is talking about.Rhett Sampson shared thisClimbing the Knowledge Pyramid: How AstraZeneca's Operations Knowledge Fabric Transforms Pharma Operations Through Connected Data Enterprise AI initiatives are failing at scale—only 4% of companies create substantial AI value despite massive investments. The core issue is context starvation: while GenAI excels at reasoning, it lacks the nuanced enterprise knowledge and semantic understanding needed to make accurate, trustworthy decisions in complex operational environments. Traditional data architectures treat knowledge graphs as glorified databases, leading to semantic drift, brittle integrations, and AI systems that hallucinate confidently but incorrectly. As organizations race toward autonomous operations and agentic AI, the ability to encode, access, and leverage enterprise context becomes the defining competitive advantage. Without semantic foundations, AI agents remain general-purpose assistants rather than specialized domain experts. The pharmaceutical industry faces unique challenges with regulatory compliance, complex supply chains, and life-critical decisions—making semantic precision not just valuable, but essential for patient safety and operational excellence. AstraZeneca's Operations Knowledge Fabric represents a fundamental architectural shift from data lakes to knowledge fabrics. They implemented a distributed semantic architecture using: * Semantic-first data products * Layered ontology model * Living ontologies * Graph-native infrastructure * AI-enabled tooling Rather than centralizing knowledge in databases, they embed semantics directly into data products and knowledge graphs, creating an interconnected web of meaning across manufacturing, supply chain, and drug development operations. Impact and Results * Accelerated AI deployment * Reduced development time * Enhanced decision quality * Future-ready architecture Full talk: https://lnkd.in/dFxRZX6m -- Maria Sorokina, PhD. Knowledge Graph and Semantics Lead, AstraZeneca Dr. Maria Sorokina is the Knowledge Graph and Semantics Lead at AstraZeneca, where she spearheads the Operations Knowledge Fabric (OKF) initiative—a semantic architecture transforming pharmaceutical operations through connected data and AI-enabled decision making. -- Connected Data London 2025 brought together leaders and innovators. Were you there? 🎥 Watch the sessions: https://lnkd.in/dh5KMvtv 📩 Join the community: https://lnkd.in/dWgyzZf Welcome to Connected Data London's #TeaserTuesday Every Tuesday, we share teasers from #CDL25 on our channels Tune in and learn from leaders and innovators; subscribe and watch premieres as they are released! Join community legends and new voices in #CDL25 for all things #KnowledgeGraph #Graph #analytics #datascience #AI #graphDB #SemTech #Ontology
-
Rhett Sampson shared thisInteresting day. Great to see Australia taking a bigger role in AI. Unfortunate timing but these things happen. SPAN tightens AI security across the board. SPAN moves security from the edges to the fabric — if it doesn’t meet policy, it doesn’t move.
-
Rhett Sampson shared thisThis is what McKinsey et al don't tell you about actually doing AI. And why you should do it and the cost of not doing it. Billiant. https://lnkd.in/gMr4UtnHThe AI-Native Playbook: How to Build, Govern and Scale the AI-Native EnterpriseThe AI-Native Playbook: How to Build, Govern and Scale the AI-Native Enterprise
-
Rhett Sampson shared thisSusan got it right. Everyone wanted to know where we'd been up til Monday! it was effectively our debut and what a debut it was. Everyone got SPAN and its potential to solve many of the problems in AI and build a $T business out of Australia. 🇦🇺 so many great contacts to follow up on. Thank You Tech Council of Australia Lucinda Longcroft Kate Cornick and the entire TCA team! We could never have done this.Rhett Sampson shared thisHello, I love this photo taken by Sheryle Moon of the Amercian Chamber of Commerce in Australia showing Rhett Sampson pitching at the Tech Council of Australia Innovation Showcase inside Parliament House. The most common response from people we spoke with: here is my business card, let's talk! We spoke with members of Parliament, advisors, public and private sector leaders and other innovators about the power of SPAN - a network that understands and coordinates intelligently, for the $Tn AI Agent Economy. SPAN solves intent, context, provenance and policy so agents work more efficiently, can scale, and save energy, water and resources. SPAN is Australian, our reach is global. Curious - connect with Rhett or I for a briefing. #Australianinnovation #innovationaus #aiinfrastructure #sovereigncapability #owningthemeansofproduction #madeinaustralia
-
Rhett Sampson shared thisAI is becoming intelligent decentralisation. Where does OpenClaw run? on your computer. Where did agents used to run? On an AI datacentre burning power. Same will happen for inference. This is a good thing. Watch this space 😜
Rhett Sampson commented on a post
15h
It's actually an OAuth issue and OAuth breaks under AI. SPAN uses defence grade cryptographic security.
Rhett Sampson replied to a comment
19h
David Barrett hey David I assume you're talking about global capacity rather than individual connections? In which case I agree 100% but SPAN addresses that by increasing efficiency.
-
Rhett Sampson liked thisLet me guess, AI was reviewing the bugs...Rhett Sampson liked thisWARNING: Lovable is experiencing a mass data leak. If you built a project before Nov 2025, your source code, database credentials & AI chat history can be viewed by any free account today. Here's how to triage ASAP 👇 Hell of way to start a week lol Let's just treat this as a fun "part two" to yesterday's fiasco... (And don't get me started on the vibe coding tax). Anyway... This is the result of a Broken Object Level Authorization (BOLA) flaw in Lovable's API. Firebase auth tokens are checked, but ownership is never verified. Any free account can query any project. Affected endpoints: /projects/{id}/*, /git/files, /git/file, /documents. All return 200 OK for pre-November 2025 projects. A researcher (weezerOSINT) created a free account and pulled the full source tree of an admin panel built for Connected Women in AI a real Danish nonprofit. 3,703 edits this year. Last touched 10 days ago. The source contained hardcoded SUPABASE_URL, SUPABASE_PUBLISHABLE_KEY, and SUPABASE_SERVICE_ROLE_KEY. The service role key bypasses every Supabase security policy. The researcher pulled real names, companies, LI profiles of speakers from Accenture Denmark and Copenhagen Business School. This was reported to Lovable on March 3 via HackerOne (report #3583821). Lovable shipped ownership checks for NEW projects and left every existing project exposed. A project created in April 2026 returns 403 Forbidden. 48 days later, the HackerOne report is still open. Per the researcher, affected accounts include employees at Nvidia, Microsoft, Uber, and Spotify. And every AI conversation (schemas, error logs, credentials, and business logic you, etc is stored and reachable through the same flaw). This is Lovable's second major security incident in 12 months. What to do: 1 / Understand what "private" actually means on Lovable Legacy projects are readable by any free account via the public API. Lovable is calling this "intentional behavior." Their recommended fix is set your project to private. That is a paid-tier feature. 2 / Rotate everything that ever touched a Lovable project built before Nov 2025 Rotate Supabase service role keys FIRST. These bypass RLS entirely Rotate every database credential (connection strings, anon keys, publishable keys) Rotate every API key you pasted into a Lovable AI chat for debugging Rotate every third-party token in your source (Stripe, SendGrid, OpenAI, auth providers) Assume customer data in your Lovable-provisioned Supabase has already been enumerated 3 / Audit your actual exposure Open your project URL in incognito. If you can load the code / chat history, so can a stranger. Check Supabase logs for anomalous read queries on user, billing, or admin tables over the last 48 days Check your Lovable project's request logs for traffic to /projects/*, /git/files, /git/file, and /documents endpoints from unfamiliar sessions If you ever pasted a credential into a Lovable chat, treat it as compromised.
-
Rhett Sampson liked thisRemember: this is an API issue more than an AI issue. If a bank leaves a sack of cash meant for the van on the sidewalk, it wasn't your PIN number that got you robbed. Lovable should've been on this: of course their biggest vendor/competitor Anthropic also had a similar hack last month, so there's that.Rhett Sampson liked thisWARNING: Lovable is experiencing a mass data leak. If you built a project before Nov 2025, your source code, database credentials & AI chat history can be viewed by any free account today. Here's how to triage ASAP 👇 Hell of way to start a week lol Let's just treat this as a fun "part two" to yesterday's fiasco... (And don't get me started on the vibe coding tax). Anyway... This is the result of a Broken Object Level Authorization (BOLA) flaw in Lovable's API. Firebase auth tokens are checked, but ownership is never verified. Any free account can query any project. Affected endpoints: /projects/{id}/*, /git/files, /git/file, /documents. All return 200 OK for pre-November 2025 projects. A researcher (weezerOSINT) created a free account and pulled the full source tree of an admin panel built for Connected Women in AI a real Danish nonprofit. 3,703 edits this year. Last touched 10 days ago. The source contained hardcoded SUPABASE_URL, SUPABASE_PUBLISHABLE_KEY, and SUPABASE_SERVICE_ROLE_KEY. The service role key bypasses every Supabase security policy. The researcher pulled real names, companies, LI profiles of speakers from Accenture Denmark and Copenhagen Business School. This was reported to Lovable on March 3 via HackerOne (report #3583821). Lovable shipped ownership checks for NEW projects and left every existing project exposed. A project created in April 2026 returns 403 Forbidden. 48 days later, the HackerOne report is still open. Per the researcher, affected accounts include employees at Nvidia, Microsoft, Uber, and Spotify. And every AI conversation (schemas, error logs, credentials, and business logic you, etc is stored and reachable through the same flaw). This is Lovable's second major security incident in 12 months. What to do: 1 / Understand what "private" actually means on Lovable Legacy projects are readable by any free account via the public API. Lovable is calling this "intentional behavior." Their recommended fix is set your project to private. That is a paid-tier feature. 2 / Rotate everything that ever touched a Lovable project built before Nov 2025 Rotate Supabase service role keys FIRST. These bypass RLS entirely Rotate every database credential (connection strings, anon keys, publishable keys) Rotate every API key you pasted into a Lovable AI chat for debugging Rotate every third-party token in your source (Stripe, SendGrid, OpenAI, auth providers) Assume customer data in your Lovable-provisioned Supabase has already been enumerated 3 / Audit your actual exposure Open your project URL in incognito. If you can load the code / chat history, so can a stranger. Check Supabase logs for anomalous read queries on user, billing, or admin tables over the last 48 days Check your Lovable project's request logs for traffic to /projects/*, /git/files, /git/file, and /documents endpoints from unfamiliar sessions If you ever pasted a credential into a Lovable chat, treat it as compromised.
-
Rhett Sampson reacted on thisRhett Sampson reacted on thisA client came to me with a data migration problem that their original provider wouldn't touch. 4,000+ entries with inconsistently inverted dates, no API pathway, and an internal estimate of 315 hours of manual correction sitting on the table. We resolved it in four hours using a browser agent. I didn't drop an AI tool on the problem and walk away. I designed the agent, defined the requirements, guard-railed potential impact and built the logic for identifying which records needed correction. The expertise is in the identification of use case and setup. That part isn't automatic. But a more important part of this story. The client has now spotted three more opportunities for this type of agent, other pockets of repetitive data work across their organisation. And we're building each one so their ops manager can run it independently. The design work happens once. The capability transfers to the team. That's the implementation model I think actually works. Not AI that keeps you dependent on a specialist. AI that gets handed over and you learn along side someone with expertise. If your organisation has work that looks like this; manual, repetitive, attention-intensive, error-prone, it's worth asking whether it's genuinely irreplaceable human work or just unexamined process. #ResponsibleAI #AgenticAI #AIImplementation #EthicAI #AIforBusiness
-
Rhett Sampson liked thisThis is what happens when you use AI agent frameworks that depend on OAuth. SPAN uses cryptographically anchored ID, permissions, policy, authorisation and domain security. No OAuth.Rhett Sampson liked thisWARNING: Vercel got hacked. If you've OAuth'd AI tools into Google, assume your source code, API keys, and customer data is being sold on the dark web today. Here's how to triage ASAP👇 The attack vector was a compromised third-party AI tool (Context . ai) with a Google Workspace OAuth app. One OAuth grant led to one Vercel employee account, which led to source code, NPM tokens, GitHub tokens, and production secrets for a platform hosting a meaningful percentage of the modern web. Here's what to do right now: 1 / Check if you're exposed (this takes maybe 60 seconds) 1. Go to admin . google . com 2. Security, then Access and data control, then API controls, then App access control, then Manage Third-Party App Access 3. Search for this OAuth Client ID: 110671459871-30f1spbu0hptbs60cb4vsmv79i7bbvqj {dot} apps {dot} googleusercontent {dot} com 4. If found, revoke and block immediately 2 / Vercel cleanup (assume compromise) Anything NOT marked as a "sensitive environment variable" should be treated as leaked. 1. Rotate every API key, DB credential, and token stored in Vercel env vars 2. Review your org's activity log for unusual deploys, invites, or token grants 3. Audit org members and remove anyone who shouldn't be there 4. Disconnect your GitHub integration, then reconnect it (invalidates stale tokens) 3 / Supply chain ShinyHunters claims to hold Vercel's NPM and GitHub tokens. If true, a poisoned release of next, @vercel/*, or any Vercel-published package is on the table. 1. Pin every Vercel-published dependency to an explicit, verified version 2. Rotate any credential that ever touched a Vercel build pipeline 3. Monitor build and deploy logs for anomalous steps Third-party OAuth apps that request scopes beyond the basics (name, email, profile pic) are a major new attack vector. Ask your Google admin to restrict "unconfigured third-party apps" to basic-info-only scopes. This won't be the last hack.
Experience & Education
-
GT Systems Australia
View Rhett’s full experience
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Licenses & Certifications
-
Commercial Mediator
Australian Commercial Dispute Centre
Volunteer Experience
-
Surf Race handicapper, chair social committe, bronze medallion patroller
North Bondi Surf Lifesaving Club
- Present 19 years 4 months
Patents
-
Media Distribution & Management System & Apparatus
AU AU2014904438
Other similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More