Co-founder @getfailsafe. Ex-CISO/eng leader @Grab, @Google. Writing on AI security, building companies, and lessons at 40.
Keeping the world safe
Joined August 2012
- Tweets13
- Following42
- Followers4
- Likes13
Wui | FailSafe retweeted
⚠️ Beware if you use a Ledger hardware wallet, especially if you bought one recently.
Based on information so far, it seems to be localized to a supply chain attack with one vendor. A small number of people probably bought fake (or tampered) Ledgers.
Ledger is one of the most secure and oldest hardware wallets in the industry. Stood the test of time. But these things happen.
I expect and know all BNB ecosystem players (and all industry) to help trace and recover the funds.
Ledger is investigating reports of loss of funds from users in South East Asia who purchased products from a reseller named CryptoBillis. As a precaution, and pending the results of our investigation, we have asked CryptoBilis to pause all sales and shipments of Ledger devices. We recommend Ledger users who purchased from this reseller in the last 90 days to not initiate set up if you have not done so yet. If you have set up your Ledger device, consider moving assets to a new Ledger signer (with new seed). We will continue to inform customers of updates as the investigation progresses.
Reach out to Ledger customer support through official channels with any questions: support.ledger.com/
Wui | FailSafe retweeted
🚨 JUST IN: Trump doubles down on $5,000 checks for U.S. adults if Republicans keep Congress, putting a $1.2 trillion stimulus proposal in play.
news.bitcoin.com/featured/tr…
Wui | FailSafe retweeted
this is pure f*cking treasure
A Stanford research group found how to orchestrate Claude Opus 5.5 and GPT-6 Astra together, and it completely breaks the resampling wall
most developers think test-time compute means running 32 blind retries on one model. you burn API tokens repeating the same 2-3 error clusters in loops
Stanford's paper (arXiv:2608.05643) splits the compute between two different brains:
> breadth search: GPT-6 Astra runs high-entropy exploration (T=0.8) to discover divergent solution paths
> step audit: Claude Opus 5.5 scans each step at T=0.0 and catches the exact logic fallacy
> local surgery: Opus rewrites only the broken step instead of dumping the whole reasoning chain
> verifier-free consensus: majority voting locks the proof without relying on drift-prone PRMs
the benchmark metrics outclass standard Best-of-N sampling:
> 91.2% recovery rate on detected reasoning flaws
> under 1.8% regressions on correct branches
> MATH500 jumps from 29.6% to 58.0% (+28.4% absolute gain)
> AMC-2023 improves by +7.5% with zero reward models
Astra explores the maze. Opus repairs the logic
you stop burning tokens on blind retries. you run verified reasoning depth
Readers added context they thought people might want to know
arXiv:2608.05643 does not orchestrate GPT-6 Astra and Claude Opus 5.5. One open-weight model generates, critiques, and corrects. MATH500 29.6% to 58.0% and the +7.5 AMC gain are for Qwen2.5-1.5B. Two coauthors are at Stanford. arxiv.org/abs/2608.05643
Wui | FailSafe retweeted
THIS IS F*CKING INSANE, I ADDED A CRAWLER TO JEV + MY GROK BOTS AND SLEPT, MY $100 RESTART IS NOW $12,161 AFTER 3 NIGHTS
everyone on my timeline built a web crawler this week
most of them just scrape headlines
a crawler is a bot that reads web pages for you
i plugged mine into the setup i already had
the crawler reads, Jev decides, my Grok bots trade
Jev is the AI that answers every question yes or no, under a second, under a cent
honestly the crawler alone is a toy, Jev is what turns reading into a trade
a coin called CRAWL launched at 10pm at $46K, by 11pm it was $2.05M
what the three of them did while i slept:
→ crawler went through 88 fresh coins, Jev kept one
→ 11pm crawler read 1,204 posts about the coin, Jev said real people, no farm
→ crawler walked 6,389 holders, Jev said dev wallet clean, tradable
→ 12:05am it dumped to $670K, crawler flagged it, Jev said buy, bots bought
→ 1:30am $2.81M, Jev said out, +$4,710
→ 2:35am bought again at $1.97M, 3:45am out at $3.54M, +$1,610
not every call worked, a push at 4:50am faded and cost $285
holding the coin from 11pm made you 29%
my bots made 145% on that chart, five green, one red
i woke up at 7am to $12,161.37
$100 → $2,197.67 → $4,962.34 → $12,161.37
at second 3 the spider taps the $670K bottom
i wrote up the whole chain, crawler to Jev to bots, it's below ↓ costs nothing
bookmark this and check me tomorrow
show this to the one friend whose crawler only scrapes headlines
night 4 is tonight, does the crawler make Jev better or is $12,161 the top?
Wui | FailSafe retweeted
Doing a bit of a self-experiment.
Goal: use my personal health and travel data to provide personalized diet and exercise recommendations for me, using frontier models but in a way that avoids leaking to them any private information.
Strategy: use a local model (Qwen 3.8 Flash Next) to orchestrate, and use powerful remote models as a tool call to benefit from higher-level thinking and knowledge that the local model does not have.
Three-layer approach to privacy:
Avoid leaking personally identifiable information, or leaking my identity through writing style -> local model writes the queries to the frontier model, not me
Avoid leaking who I am through the payment channel -> zkAPI
Avoid leaking who I am through networking / IP -> Tor
You need all three (and finally we have all three, at least to some extent)
A skill file teaches the local model when and how to construct minimally-data-revealing requests to remote models. Use zkAPI-wrapped-with-Tor as a CLI tool.
And everything works! I got the recommendations back, info from frontier models helped to improve them.
Main deficiencies:
* Tor is really not optimized for request-by-request de-linking, which is the only form of network-layer privacy that really makes sense in today's world (long-running identifiers are too fragile). Probably not private enough, and latency 10-100x higher than it could be, at the same time.
* The skill file's request construction strategies are definitely far from optimal.
* Qwen 3.8 Flash Next is still too slow for comfort. It's comfortably running at 20-30 TPS, but it would only really start to feel fast at 100+
* There is a tradeoff: the more careful you are about what data you give to a remote model, the less it can help you
github.com/ethereum/zkapi/pu…
Wui | FailSafe retweeted
I built an AI Influencer automation in Arcads
... that Automatically Create UGC Videos while you sleep
Comment “UGC” and I’ll send you the full workflow 👀
Wui | FailSafe retweeted
THIS GUY MAKES $1,000 A DAY BY WAITING FOR THE LAST FEW SECONDS ON POLYMARKET.
No predictions, no drama, just buying 99 cent contracts when the outcome is already basically decided.
Wui | FailSafe retweeted
be a north korean hacker:
> spend years at secret hacker school
> get sent to China to learn the free internet
> return as a state sponsored cyber soldier
> pose as recruiter and dm devs
> promise $300K remote job
> get their resume
> mass apply for a remote IT job
> have 10 interviews per day
> spend months looking normal
> let team download malicious stuff
> compromise the platform
> bridge via Tornado Cash
> cashout via non KYC chinese exchange
Wui | FailSafe retweeted
A Chinese engineering student walked into his dorm with three boxes from Apple. Two Mac Studios. One Mac Mini. Set them up in a row. Stuck a label on each one: UI/UX. DEV. ADMIN. Two monitors behind them showing satellite maps of cities he's never been to.
His roommate watched the whole thing and said: You spent $4,000 to check the weather?
He smiled. Said nothing. Let him keep thinking that.
Then someone noticed what the three boxes were actually connected to.
A wallet. Making $106K. Betting on the temperature.
ColdMath. $106,875 profit. 5,623 predictions. Joined November 2025. Bio: Edge Compounds.
→ polymarket.com/@ColdMath?via…
Two Mac Studios and a Mac Mini doing one thing. Claude pulls live pilot weather data. METAR. TAF. Real sensors from real stations. Updated every 1-3 hours worldwide. Temperature accurate to a tenth of a degree. The DEV box compares it to prediction market prices. When they don't match the UI/UX screen flashes. The ADMIN box logs the trade.
Flash. Trade. Green.
$13 on Lucknow hitting 39C on March 7. Payout: $6,850. $11 on Ankara hitting 4C on February 23. Payout: $5,752.
Eleven dollar bets returning five thousand. On the temperature in a city most people can't find on a map.
A friend who flies commercial told him pilots get atmospheric data hours before any public forecast. This data is free. Aviation safety requires it. Nobody outside of aviation even looks at it.
He looked. Pointed Claude at the feeds. Said: find me every city where the real temperature doesn't match the price.
Claude found dozens. Every single day. Tokyo. Chicago. Wellington. Atlanta. Ankara. Lucknow. Cities on six continents. All with weather stations publishing data that nobody in the markets is reading.
The three boxes run 24/7. Even when he's in class. Even when he's asleep. The satellite maps keep updating. The DEV box keeps comparing. The screen keeps flashing.
His roommate finally asked what the setup actually does. The student showed him the balance. The roommate didn't say anything. Just asked for a third monitor.
34K people watching. $89K still loaded in active positions. Two Mac Studios. One Mac Mini. Two screens. One quiet kid who realized the most predictable thing on Earth is the thing everyone ignores.
The weather.
🤝 Paid partnership
Wui | FailSafe retweeted
🚨 CRYPTO SECURITY ALERT: THE END OF FACIAL VERIFICATION (KYC) 🚨
🌐 The launch of JINKUSU CAM—a cybercriminal tool—has been detected. It is a powerful AI suite designed specifically to BREACH the security protocols of the world's largest exchanges (Binance, Coinbase, Kraken, OKX).
👤 Developer/Threat Actor: jinkusu.
🛠️ Tool Type: Real-time media manipulation software (Live Deepfake).
🎯 Objective: To bypass KYC (Know Your Customer) protocols on financial platforms, cryptocurrency exchanges, and mobile banking applications.
📦 TECHNICAL FEATURES (ATTACK VECTORS):
The software utilizes cutting-edge AI technologies to deceive verification systems:
🎭 Real-time Face Swap: GPU-accelerated face replacement (CUDA/DirectML) using InsightFace for fluid gesture transfer.
🗣️ Voice Changer: Real-time voice modulation with pitch adjustments and preset profiles (Anonymous, Radio, Robot) to evade voice biometrics.
🎥 Virtual Camera: Output compatible with OBS Virtual Camera, allowing the manipulated video feed to be injected into Zoom, Teams, Chrome, and verification apps.
📱 Emulator Support: Designed to function within Android emulators, enabling attacks against mobile applications that require "live" selfies.
✨ AI Enhancement: Utilizes GFPGAN and 478-point facial meshes (MediaPipe) to ensure the fabricated face mimics human expressions with extreme precision.
⚠️ ASSOCIATED RISKS (CRITICAL):
🏦 Mass Banking Fraud: Enables criminals to use stolen photos (such as those from previous data leaks in France or Mexico) to create a "living" persona that speaks and moves convincingly before a bank's camera.
🎭 Synthetic Identity Theft: Facilitates the creation of highly convincing fake identities for money laundering activities and romance scams (pig butchering). 🔓 KYC Compromise: Bypasses the "Liveness Detection" checks that many applications consider secure.
#Cybersecurity #Deepfake #KYC #Bypass #JinkusuCam #AI #IdentityTheft #Fintech #InfoSec #CyberAlert
