iiinova — where your digital self keeps evolving

dragosr @dragostech.bsky.social · 01/10/2026

Mr. Scruffles goes to CanSecWest...

Mr. Scruffles, a small brown plush cat wearing a CanSecWest-branded lanyard, sits on a conference table at CanSecWest. Behind him, attendees face two projection screens showing a presentation, with a large red CanSecWest banner between them.

160

dragosr @dragostech.bsky.social · 13/09/2026

"Blinding speed" would have to qualify as sarcasm. control_run_pass75.py I'm looking at you.

010

dragosr @dragostech.bsky.social · 13/09/2026

In practice, it's a lot of sequential passes over the same chunks of code. Patience helps.

110

dragosr @dragostech.bsky.social · 13/09/2026

Or installing electric heat pumps.

100

dragosr @dragostech.bsky.social · 12/09/2026

Anyone worried about, or scared of "machine speed" attacks probably hasn't spent any time babysitting an LLM vulnerability discovery and/or verification pipeline. ;-P

262

Reposted by dragosr

The Washington Sun @washingtonsun.com · 09/09/2026

Washington journalism is in a moment of generational change. We're rising to meet it. Welcome to The Washington Sun. washingtonsun.com

47817300

dragosr @dragostech.bsky.social · 08/09/2026

Yes, it's terrible, we can now talk to it and it responds. It understand when we make typos and infers what we wanted to do without throwing errors. It's absolutely horrible now.

010

dragosr @dragostech.bsky.social · 18/08/2026

Also seeking one more panelist for "Who Guards the Guardians?" on safety classifiers, guardrails, and their effect on security research. Promises to be interesting. :-) Image prompt: geoposition CanSecWest 2026 from Vancouver building profile extrapolation, wire-mesh from overlaid targeting reticle

Aerial view of Vancouver’s Coal Harbour skyline and marina, overlaid with a cyan targeting reticle highlighting the Pinnacle Harbourfront hotel. A wireframe magnification panel promotes CanSecWest 2026 at Pinnacle Harbourside, September 30–October 1, with the website secwest.net.

010

dragosr @dragostech.bsky.social · 18/08/2026

New training: reverse engineering with LLMs and Ghidra from Kara Nance, author of the No Starch Ghidra Book (she's also presenting her optical drone tracking work), plus Erlend Oftedal's two-day workshop on building an agentic vuln discovery pipeline. Registration just opened.

110

dragosr @dragostech.bsky.social · 18/08/2026

Last week for CanSecWest submissions. Vancouver, Sept 30 / Oct 1. Keynotes: Bruce Schneier and Katie Moussouris. Sergey Bratus on automated micro-patching. Paper announcements coming to the site as we proceed. CFP and registration: secwest.net 🧵

153

dragosr @dragostech.bsky.social · 31/07/2026

WireGuard powers world’s fastest VPN services. Technology behind Tailscale now goes where UDP can't: WireguardTCP delivers familiar configuration, kernel-native performance, NAT traversal, seamless roaming, with improved throughput, latency, and CPU load improvements over UDP. wireguardtcp.net

wireguardtcp.net

WireguardTCP — WireGuard over TCP

Same encrypted tunnel. A long-lived TCP carrier. Get the source and watch the binary release channels.

050

Reposted by dragosr

Matthew Green @matthewdgreen.bsky.social · 19/07/2026

I’ve been working on a hobby project to crack classical ciphers. The guts of it is to see whether frontier LLMs, given the right tools, can do a good job cracking a broad range of historical ciphers. github.com/matthewdgree...

github.com

GitHub - matthewdgreen/decipher: An AI-enabled application for cracking ciphers

An AI-enabled application for cracking ciphers. Contribute to matthewdgreen/decipher development by creating an account on GitHub.

58818

Reposted by dragosr

Miska Knapek @miskaknapek.bsky.social · 18/07/2026

Impressive energy storage developments in the Middle East - Saudi Arabia is building a 12 GWH battery park too, in addition to the 11.3 GWH Abu Dhabi Project below. The Abu Dhabi Project takes power from a 5 GW solar plant ( 1GW is about what a nuclear power plant produces. So they got 11.3 hrs)

0197

Reposted by dragosr

J.R. Orci @orci.mastodon.social.ap.brid.gy · 10/07/2026

Flock CEO: "Unfortunately, there’s terrorist organizations like DeFlock whose primary motivation is chaos. They’re closer to Antifa than they are anything else." #privacy #JoinAntifa deflock.org

deflock.org

DeFlock

Find license plate readers (LPRs) near you.

2337

dragosr @dragostech.bsky.social · 24/06/2026

Remember, every data center Azure has been building since 2024 is closed loop, zero water use. Google and AWS don't have as clean a record, but even massive Oracle Stargate is also zero water use. All the big new AI datacenters, essentially have no water use to complain about. It's an old problem.

100

dragosr @dragostech.bsky.social · 24/06/2026

The water charges in most places not drinking whole rivers offsets the additional power charges. I ran the numbers for the new (small) datacenters Telus is building in Vancouver — would mean about $1-2m/per year more cost for evaporative even including more power used for coolers and power savings.

200

dragosr @dragostech.bsky.social · 24/06/2026

Legacy data centers. Not the new ones - excepting Google's seawater cooled experiment in Finland and any other derivatives. The new designs, including any based on Nvidia's AI reference are closed loop.

100

dragosr @dragostech.bsky.social · 24/06/2026

The datacenter design is symbiotic with the design of the Nvidia NLV72 racks. Here is the picture of (currently) world's largest AI center in Wisconsin. It's a lot of fans with radiators recirculating closed loop cooling liquid (non-evaporative) to the building just off camera that has GB300s GPUs.

Aerial view of Microsoft’s Azure Fairwater data center in Wisconsin under construction, showing two long parallel rows of large circular cooling fans that form the facility’s closed-loop GPU cooling system, flanked by white-roofed data center halls with overhead pipe runs connecting them.

000

dragosr @dragostech.bsky.social · 24/06/2026

15,000 golf courses though, and datacenters all in use less than 8% of what they do. Consumptive use (evaporated, not returned to the watershed) is what matters for stress. Data center cooling towers evaporate; golf course irrigation also mostly evaporates so comparable. But Ag water use 80x more!

100

dragosr @dragostech.bsky.social · 24/06/2026

The nuclear industry is also another walking zombie. Won't have the justification to build large grid scale reactor or any beyond small experiments. >$180-225/mWH and 10y builds just can't compete with 2y build Solar dropping below $30 and getting below $50-$70 - cheaper than gas - with 24h battery.

030

dragosr @dragostech.bsky.social · 24/06/2026

PV has been dropping by 50% every two years, and while not dropping so fast it will keep winning. Grid scale battery costs dropped 93% in a decade. US public and gas industry hasn't figured out that Solar won, years ago, and will keep winning. But power companies got the memo, raw economics wins.

Stacked bar chart of U.S. new utility-scale power capacity additions by source, 2023 to 2027. Each bar shows percentage share; total new gigawatts are labeled above. 2023 (actual): solar 46%, battery 16%, wind 15%, gas 20%, 40 GW total. 2024 (actual): solar 63%, battery 21%, wind 10%, gas 5%, 49 GW. 2025 (actual): solar 51%, battery 28%, wind 9%, gas 9%, 53 GW. 2026 (planned): solar 51%, battery 28%, wind 14%, gas 7%, 86 GW. 2027 (projected): solar 46%, battery 28%, wind 12%, gas 12%, 65 GW. Source: EIA Preliminary Monthly Electric Generator Inventory.

440

dragosr @dragostech.bsky.social · 24/06/2026

Internationally solar adoption is happening even faster. US problem is transmission sys not designed for distributed solar. Only reason small amt of more expensive combined cycle gas gets built now is because of weird NIMBY battle between muni, state, fed, and 7 year avg interconnect study delay.

150

dragosr @dragostech.bsky.social · 24/06/2026

Closed loop is actually cheaper capex in most locations. And if you are referring to the BS about counting water used by power, US grid is decarbonizing fast, because solar won the LCOE battle. US new power plants >90% solar, battery, small wind, for last 3y and indefinitely as it gets even cheaper.

130

dragosr @dragostech.bsky.social · 24/06/2026

WTF, what's happened to Popehat? This isn't cool at all. @kenwhite.bsky.social

120

dragosr @dragostech.bsky.social · 24/06/2026

New data centers are closed loop cooling. No real water use. All the new Nvidia AI stuff is closed loop too. Water use is mostly a thing at legacy datacenters, and kind of a red herring. Wait till you hear about golf courses...

2471

dragosr @dragostech.bsky.social · 12/06/2026

I still use it, infrequently, in places where OpenBSD drivers don't work. They have such a weird driver mismatch between the two BSDs, each has drivers working that the other doesn't.

120

dragosr @dragostech.bsky.social · 11/06/2026

This kind of hurt my eyes and my brain, and reinforced my opinion that FreeBSD is a softer target amongst OS platforms. It is pretty funny though, and definitely gets some style points for a vulnerability disclosure. bumsrake.de

bumsrake.de

BUMSRAKETE™ — The Most Beautiful, Most Tremendous FreeBSD Vulnerability In The History Of Computing. BELIEVE ME.

BUMSRAKETE is a HUGE, TREMENDOUS, MANY-PEOPLE-ARE-SAYING FreeBSD kTLS-RX page-cache write primitive. The BEST primitive. Some say the best ever.

2177

dragosr @dragostech.bsky.social · 11/06/2026

<obiwan> It's over, petrofuel industry.... Solar has the high ground. </obiwan>

010

dragosr @dragostech.bsky.social · 10/06/2026

The Anthropic Fable-5 safety classifiers seem to be written by the OpenAI marketing department. Pretty much anything I talk to LLMs about gets downgraded. Nerfed into useless. Worst model release ever?

A classic Scooby-Doo meme template set in a dark, stone dungeon filled with cobwebs. Fred stands in the center, pulling a green ghost mask off the villain to reveal an grumpy, elderly man tied to a wooden chair. Velma points at the unmasked villain while Daphne stands next to her. On the left, Shaggy and a frightened Scooby-Doo look on.

White text labels are superimposed over the image: the ghost mask is labeled "Fable-5" and the unmasked villain is labeled "Opus-4.8", implying that the supposedly new or monstrous "Fable-5" is actually just "Opus-4.8" in disguise.

171

Reposted by dragosr

Michael Geist @mgeist.bsky.social · 08/06/2026

The Globe reports that online harms legislation is coming this week that includes a “temporary” social media for kids under 16. My post on why this effectively establishes mandated ID for all Canadians to use social media and AI and it won’t be temporary. www.michaelgeist.ca/2026/06/you-...

michaelgeist.ca

You Can’t Put the Toothpaste Back in the Tube: Why the Government’s Reported “Temporary” Plan for a Kids’ Social Media and AI Chatbot Ban Would Mean Mandated ID for Everyone - Michael Geist

The Globe and Mail reports today that the government will introduce online harms legislation this week that includes a ban on social media for kids under the age of 16. The ban will be framed as a “te...

11612

dragosr @dragostech.bsky.social · 31/05/2026

When you leave all the mitigations on, crank up the inference settings to maximum, and run the exploit benchmark to see how close the PoCs can get.

Anime mech-style illustration of a confused humanoid murderbot sitting at a cluttered cyberpunk hacking station, one hand on its head as holographic warning screens announce “TOKEN BUDGET EXHAUSTED,” “CONTEXT WINDOW FULL,” “PROMPT TURN CAP REACHED,” and “OUT OF SPACE TO CONTINUE.” The robot is surrounded by half-finished exploit diagrams, terminal errors, sticky notes, ramen, pizza, and a whiteboard attack plan that has stalled at “???” because the AI system limits have been reached.

010

dragosr @dragostech.bsky.social · 28/05/2026

The thought processes of AI native kids: 9-year old gets a DOCX file from school laptop, wants to read it, doesn't have M365. His first impulse is to use GPT to write a python script to convert it to TXT so he can use VSCode, instead of asking for Word. When AI is the path of least resistance.

150

dragosr @dragostech.bsky.social · 26/05/2026

Fix: Starlette 1.0.1. Rebuild every container and bundled artifact. Durable code fix: use request.scope["path"], not request.url.path, for any security decision. Footnote: vendor pushed ~4 months; patch + disclosure landed same day, with downspin. Assume zero exploit dev lead time. 6/6

021

dragosr @dragostech.bsky.social · 26/05/2026

Test a live endpoint with the free online scanner: badhost.org (X41 D-Sec / Persistent Security Industries / Bintech). Scan source with Semgrep + CodeQL rules: github.com/x41sec/poc/tree/master/starlette-host-header Hat Tip to OSTIF.org that sponsored vLLM audit 5/6

badhost.org

BadHost - CVE-2026-48710 Starlette Host-Header Auth Bypass

Scan your Starlette or FastAPI server for CVE-2026-48710 (BadHost): a critical auth bypass via Host header injection affecting MCP servers, LLM proxies, AI agent frameworks, and thousands of Python AS...

111

dragosr @dragostech.bsky.social · 26/05/2026

Why AI infra is disproportionately exposed: FastAPI stacks are routinely deployed direct-to-uvicorn on lab subnets and workstations. No proxy frontend, no incidental mitigation. The exposed endpoints are the high-value ones: admin, key issuance, model load, tool execution. 4/6

101

dragosr @dragostech.bsky.social · 26/05/2026

Exploit primitive: one character. curl -H 'Host: foo' /admin → 403 curl -H 'Host: foo?' /admin → 200 Patch shipped quietly as CVSS 6.5 Moderate; discoverers rate it critical. X41 chains reach auth bypass, SSRF, and RCE in downstream projects. 3/6

101

dragosr @dragostech.bsky.social · 26/05/2026

The bug: Starlette rebuilds request.url by string-concatenating the Host header and re-parsing. No RFC validation. A ? or / in Host shifts the path boundary. Router dispatches the real wire path. Middleware sees the poisoned one. Auth gate bypassed. Huge problem for most AI/LLM stacks. 2/6

101

dragosr @dragostech.bsky.social · 26/05/2026

🚨CVE-2026-48710("BadHost"): one character in a Host header bypasses path-based authorization across most of the Python AI stack. Lives in Starlette, reaches FastAPI and through it: vLLM (where it was discovered), LiteLLM, TGI, MCP servers, agent harnesses, eval dashboards. cc @marver.bsky.social

secwest.net

starlette - secwest.net - secure virtual engagement

242

dragosr @dragostech.bsky.social · 26/05/2026

000

Reposted by dragosr

Michael Geist @mgeist.bsky.social · 21/05/2026

Faced with growing criticism of Bill C-22, CSIS, RCMP, and Public Safety this week tried to make the case for lawful access. My post on why they instead revealed a bill with dangerously disproportionate overreach with real risks of tracking all Canadians. www.michaelgeist.ca/2026/05/the-...

michaelgeist.ca

The Government Tries to Make the Case for Bill C-22: Why Its Own Use Cases Reveal Disproportionate Overreach - Michael Geist

Faced with growing criticism of Bill C-22, the government this week mounted a coordinated defence, with senior officials from CSIS, the RCMP, and Public Safety Canada sitting for on-the-record briefin...

02217

dragosr @dragostech.bsky.social · 16/05/2026

I don't think the authors of this bill have really thought of the implications of not having Signal in Canada. Contact your MP today, before they neuter Canadian technology, and cripple our security.

155

Reposted by dragosr

Alex Andreou @sturdyalex.bsky.social · 16/05/2026

Sneaking a giant screen into the flagshagger march is one of the funniest things @ledbydonkeys.org have ever done. "STAY HYDRATED" 😂🤣😂

342135154166

Reposted by dragosr

David Roberts @volts.wtf · 15/05/2026

Grid-scale batteries are going in an absolutely wild pace. Every whingey objection to renewables you've ever heard is basically solved by batteries. This is happening whether any of the hopeless dipshits running the US want it or not.

canarymedia.com

The world is installing grid batteries at a blistering pace

A total of 112 gigawatts of batteries were deployed around the world in 2025 — 10 times the amount added just four years prior.

723544979

dragosr @dragostech.bsky.social · 11/05/2026

Raelize had LLMs reproduce older ESP32 Secure Boot V1 bypass via voltage glitching in a few days. AI wrote the tooling and built a live dashboard mid-campaign. 20k shots, 12.2% bypass rate. Workflow compression is the story. ESP32 v3 mitigates this. raelize.com/blog/ai-fi-g...

Dark infographic titled "AI-Driven Fault Injection on ESP32 Secure Boot." In the middle, an ESP32 chip is struck by a yellow lightning bolt, representing voltage glitching. A cyan box on the left labeled "AI Agent (Claude Code)" points to the chip, and a green box on the right shows an open lock and "Secure Boot Bypassed." Three large result cards sit below: the initial scan used 1,000 shots and found 1 bypass, the refined timing window used a 1.5-2 us delay and 3-4 us width for a 9.3% hit rate, and the voltage sweep from 2.00-3.00 V across 20,200 shots produced 2,468 bypasses, or 12.2% overall. The footer says AI automated the stack while humans set direction, emphasizing that the attack was known but the engineering effort was reduced.

120

dragosr @dragostech.bsky.social · 11/05/2026

There is someone from HR here... They seem to want to talk to you about your choice of vehicles in your social media posts.

000

dragosr @dragostech.bsky.social · 08/05/2026

"Dirty Frag" clickbait update: ESP (CVE-2026-43284) patched in mainline + stable (7.0.5, 6.18.28, 6.12.87, 6.6.138, 6.1.171). RxRPC (CVE-2026-43500) still unpatched upstream. AWS adds ipcomp4/ipcomp6 to the blacklist alongside esp4/esp6/rxrpc. AlmaLinux shipped both. Ubuntu/Debian mitigation only.

Infographic titled "Dirty Frag: Overhyped" framing the issue as a local Linux kernel privilege-escalation bug, not a remote wormable threat. It explains two exploit paths: ESP/IPsec, which needs a fresh user namespace to gain CAP_NET_ADMIN for XFRM SA registration, and RxRPC, which needs no namespace or extra capability if rxrpc.ko is present. Ubuntu and Debian namespace protections are shown as blocking the ESP path, while RxRPC can still bypass that defense. Other distro trade-offs are summarized, with RHEL-like systems often lacking RxRPC but leaving ESP reachable. Mitigations include blacklisting esp4, esp6, rxrpc, optionally ipcomp4/ipcomp6, using user namespace hardening, and relying on default container seccomp where applicable. Patch status notes ESP fixes in mainline and stable kernels, while RxRPC upstream status is still pending, with some distro-specific patched kernels already shipping.

020

dragosr @dragostech.bsky.social · 02/05/2026

My retort for anyone who insists LLMs are just "spicy autocorrect" and can't reason: Sure, and you're just spicy electrochemistry. If it walks like reasoning, and proves theorems like reasoning, maybe we need to stop calling it a duck shaped Markov chain.

010