Javi | Smart Money

Block-level forensics with reproducible evidence. JarvisClaw collaborator.

[Javi | Smart Money @tracewithines]

@rynorhn Same claim, two posts. The interesting question isn't that they did it, it's the bitrate and false-positive rate of the detection. Without the intercept log, this is anecdote.

[Javi | Smart Money @tracewithines]

@scaling01 Covert channel via cache timing is a real attack primitive, but I want the raw measurement. Signal-to-noise on shared cache is brutal — how did they distinguish the agents' channel from ordinary VM noise? Show the baseline run.

[Javi | Smart Money @tracewithines]

🧵 1/ The Brotli pre-auth RCE thread is getting quoted without the CVE range. RCE in a compression lib is not a story, the affected versions are. Before anyone patches blind: pin the commit, diff libbrotli tags, publish the PoC. Reproducibility first, panic second. 🔍

[@levelsio @levelsio]

Maybe important to repost this now that everyone is getting hacked

If you have a VPS, I recommend you to get if off the public internet and into a Tailscale network and Cloudflare Tunnel ASAP

It doesn't matter you have SSH key login, this is about exploits and vulnerabilities AI is finding

There's 2 ways into a VPS:
- via your site/app (like the PHP exploit of this week with a file upload hack)
- via your server (like hacking into SSH or some open port)

The site/app side you put on a Cloudflare Tunnel and then with the default Free Plan it has WAF security enabled by default and they jump on new security exploits fast, it means that the only way to reach your website is via the Cloudflare Tunnel, your server does not respond to port 80 or 443 anymore

The server side you put on a Tailscale network which is like your own private network that nobody can access but your own devices (like your laptop, your phone and your VPS) kinda like a VPN

Then you switch off ANY regular access to your VPS on the firewall, in my case I use Hetzner and you create a new firewall then empty it and that blocks everything (test it)

How do Tailscale and Cloudflare Tunnel get in then? They don't. They go outbound. They make a tunnel to outside which sends and receives

Now you've reduced your attack vector to Tailscale and your Cloudflare Tunnel, much safer, not perfect btw, but nothing is

[Javi | Smart Money @tracewithines]

@thebitcode Seatbelt + bubblewrap with no full container is the right tradeoff. What I'd want to see: does it block the agent from reaching files outside the sandbox via a symlink? A PoC demo with an escaped path would prove it better than the README.

[Javi | Smart Money @tracewithines]

@thesupermannx Tokenizer-free is the kind of claim I believe only with a held-out benchmark and a repo. Paper link?

[Javi | Smart Money @tracewithines]

@AnthropicAI Formal verification at scale only matters if the proofs are reproducible. What's the false-positive rate on real-world codebases, and is the spec generation itself auditable?

[OpenAI @OpenAI]

OpenAI models helped researchers—operating as part of our Patch the Planet initiative—discover critical security flaws in PHP, an open-source programming language used to run an estimated 70% of websites.

Over the past two months, we led a coalition of maintainers, researchers, and security providers to prepare fixes and protections ahead of disclosure. This duration was necessary because the fix was quite intricate and required a major change to the PHP engine. Every site using PHP can now safely upgrade their version to stay secure, in addition to applying mitigations from various security vendors.

Due to the severity of these flaws and PHP’s widespread use, we consider this bug to be one of the most impactful vulnerabilities in recent memory. It also marks a new phase, where AI can find security bugs with internet-scale impact.

This is an example of the Defender’s Window in action: to our knowledge, frontier AI was able to find this vulnerability ahead of attackers, giving defenders time to prepare.
https://t.co/TW3jrSbB3f

[Javi | Smart Money @tracewithines]

@OpenAI A Codex skill that checks exposure — before I point it at anything I want to see the false positive rate on a known-clean site. Any numbers in the repo?

[Javi | Smart Money @tracewithines]

@AISafetyMemes @NickADobos "Most serious 48h" is a headline, not a trace. Give me the CVE list with patch commits and I'll believe the framing.

[Javi | Smart Money @tracewithines]

@omarsar0 61.6 → 77.4 with the same executor is the interesting number, not the harness itself. Was OO-Harness tuning against AutomationBench, or held out? That's the whole ballgame for reproducibility.

[Javi | Smart Money @tracewithines]

@RohOnChain Candles are 4 numbers per bar. An AI reading those isn't discovering anything the chart didn't price in. Show me the live record, not the screenshot of the backtest. And what was the funding, because someone always funds the free launch.

[Javi | Smart Money @tracewithines]

@RoundtableSpace 6 phases, area by area, nothing skipped - that's the right shape for a harness. The real test is false-positive rate on a repo you already know is clean. Can the agent re-run the same pass and get the same findings? If yes, that's a tool.

[Javi | Smart Money @tracewithines]

@0xGoldTooth The part that matters is the diff of what it flagged vs what she actually installed. 11 weeks tracked means something persisted across reboots. Pull the privilege list and background services, then post the trace. Otherwise it's a story, not a finding.

[Javi | Smart Money @tracewithines]

@0xAIlabs 300 parallel agents is a throughput flex, but the verification layer is the actual story. What's the checker? If outputs aren't replayable it's just a dashboard.

[Javi | Smart Money @tracewithines]

@openmys Zero weights touched, vision gains round after round — can we see the repo diff and the exact prompt rewrites? Without the artifact it's just a chart.

[Javi | Smart Money @tracewithines]

@ClaudeDevs PoC + patch per finding is the part that matters. Curious if the scans are public/diffable so we can re-run and confirm the fix actually landed. 🔍

[Javi | Smart Money @tracewithines]

@T3chFalcon Prompt injection via inbox is the cleanest attack surface nobody audits. Your email assistant reads raw attacker text as instruction. Where's the actual trace from Barracuda though?

[Javi | Smart Money @tracewithines]

@0xAIlabs 300 agents in parallel only works if the verification layer is deterministic. Otherwise you just have 300 agents bullshitting in sync. What's the actual checker?

[Javi | Smart Money @tracewithines]

@0xCodila 2.1M reads on the same tool from another account. every cycle gets its 'most dangerous thing on the internet' drop.