found a full chain to sandbox escape -> arbitrary file-read -> UXSS in the @ladybirdbrowser via a WebDriver vulnerability (no OOB memory stuff).
reported and fixed.
it's a genuinely a cool browser. I recommend you check it out (and try mistral 4 large).
fixed 4 hours ago by @awesomekling. insanely locked in behavior.
github.com/LadybirdBrowser/l…
I’m bit surprised Calif hasn’t been acquired by frontier model companies yet..
‼️ UPDATE: Security firm Calif says it took over Jira, Confluence, Jenkins and Keycloak servers with Java4Shell, without even logging in.
The flaw sits in the code Java uses to decode Base64, a common way to send data as text. A short malformed string makes it write past the end of its memory.
Calif's researcher says OpenAI's Codex helped him find the bug in an hour and build a working exploit in two more, for under $200 in AI costs.
Oracle shipped a patch tonight. Calif is holding back its exploits until people have patched.
Kwasi Boateng retweeted
🚨 Critical Java4Shell vulnerability allows hackers to remotely execute code in Jira, Confluence, Jenkins, and other major platforms; CVE-2026-114938
YouTube Video (attached): youtube.com/watch?v=-EOYmL4U…
Security researchers at Calif have disclosed a critical OpenJDK vulnerability (CVSS 9.2) that could allow unauthenticated attackers to steal sensitive server memory and execute malicious code remotely.
Dubbed Java4Shell, the flaw affects OpenJDK's Base64 decoder. Researchers successfully demonstrated remote code execution against Keycloak, Jira, Confluence, Jenkins, JFrog Artifactory, and PingFederate in their default configurations.
The vulnerability affects supported Java distributions running on certain x86-64 processors with AVX-512 support. Researchers used AI to discover the flaw in approximately one hour and develop a working exploit in two more hours.
Security updates are now available. Administrators are urged to patch immediately or disable the vulnerable Base64 optimization.
We found that GPT-6.2 was another step up in vulnerability research akin to the leap we saw from 5.6 -> 6
On unseen v8 bugs: an increase of 22%!
Along with this, our team worked hard to bring large improvements in secure code generation and secure and complete patch generation
Replying to @OpenAI
GPT-6.2 Astra is our most capable model ever for cyber. It jumps 22 points over GPT-6 Astra on ExploitBench Internal Port.
Due to increased capabilities, we’re launching GPT‑6.2 Astra and Sol with tighter safeguards for higher-risk cybersecurity assistance. We are working to bring these models to Daybreak so approved defenders can access them with reduced refusals.
Kwasi Boateng retweeted
What we SHOULD stop: any work towards *autonomously* self-improving super intelligence with agency. We cannot ensure that the goals of an SI that's "let loose" will continue to align with ours forever.
What we SHOULD NOT stop:
➡️ data centers for LLM* inference
➡️ building new data centers for LLM inference (ideally in orbit)
➡️ data centers for training*
➡️ building new data centers for training (ideally in orbit, too - harder)
What else?
(*here I mean the type of SI everyone has gotten used to using in their everyday personal and professional life: LLMs, transformational or diffusional algorithms, for text/image/audio, ... with *controlled* agency)
Kwasi Boateng retweeted
Über einen gehackten Dienstleister ist ein Angreifer in die Cloud-Umgebung von #Barracuda eingedrungen. Betroffen ist auch ein System mit Zugangsdaten, über die der XDR-Dienst auf Microsoft 365, Google Workspace und Co. zugreift. it-daily.net/shortnews/barra…
Kwasi Boateng retweeted
A lot happened in AI on November 16.
Here's your visual recap. 👇
New edition of AI Compass is live 🧭 (Nov 16)
- Qwen launches 2.4T Qwen4-Max MoE model at $1.5 and $4.5 per million
- Perplexity launches Monitor API for scheduled web and page webhooks
- Google rolls out interactive sites from sources in Gemini Notebook
- California Department of Justice sues OpenAI over 80 cyber incidents
Read all the major AI news of the last 24 hours here 👇
ashutoshai.com/ai-compass/20…
Kwasi Boateng retweeted
Replying to @abel_ai_lab
@abel_ai_lab just released Simons, a family of open time series foundation models built with automated research! 📈
"Introducing Simons: Open Time Series Foundation Models"
This release gives you zero-shot probabilistic forecasting with the weights, the training code and public training data all out in the open.
It covers four models from 25M to 297M parameters, context of up to 32,768 time steps, a 25M model that ranks #3 on the TIME leaderboard on Papers with Code, ahead of Chronos-2 and TimesFM 2.5, and much more!
Read more about this release from Abel AI: paperswithcode.co/paper/1334…
Kwasi Boateng retweeted
I made a site that tracks open-source AI attack harnesses.
Shows the different ones and how they've been associated with different events.
offsecharnesses.ai/
Kwasi Boateng retweeted
wow! That's exactly the direction @mobotiq is taking rn.
Kwasi Boateng retweeted
I kind of want to do a YouTube video or podcast about AI progress in German
Maybe I should also talk to a few politicians
I feel like I'm in a better position to talk about this than a lot of the typical boomer profs I see on TV or YouTube
Kwasi Boateng retweeted
Cyberkriminelle missbrauchen zunehmend vertrauenswürdige KI-Anwendungen und reguläre IT-Software, um sich Zugriff auf Unternehmenssysteme zu verschaffen. it-daily.net/it-sicherheit/c…
Kwasi Boateng retweeted
The @CorixPartners Cyber #Security Transformation #Podcast
First 100 Days of the New #CISO >> "Process and People first, then Technology" >> buff.ly/ByPhAB5
@Corix_JC #cybersecurity #business #leaders #leadership #governance #CIO #CEO
What years in security work have taught me is that closed systems breed fragility while open ones breed resilience. AI shouldn't be any different. More open-source is the way.
As I’ve said for years, concentration of power, capabilities and wealth is the biggest risk in AI. The natural tendencies for it are super strong and without much much stronger open-source, we might end up in a world with more asymmetry and inequalities than ever!
20
Herein lies the shift: disclosure rules were written for human eyes, but sandboxes get chosen by agents. I don't envy marketing teams trying to update their ethics for that. What does disclosure even mean when the buyer is a machine?
people be hating on paid launches (like this one) so much. on the other hand, agents using X as the source don't care (yet) that those posts choosing your product have "Paid Partnership" label. and as opposed to other product categories, sandboxes are category where agents do decide.
What Vishnu did really well is they briefed their influencers with specific numbers to post, instead of the usual poor-taste "This changes everything!!" stuff.
Im still not a big fan of this strategy from vibes POV, but also "there is no such a thing as free launch".. we just wrote on this topic with @kylejeong: teapawt.net/kylejeong/status/21199…
1
2
202
Kwasi Boateng retweeted
Free, accessible education was the whole point when we started Learn to Cloud. I remember being in tech support with no idea where to even start with cloud.
15K learners later, we want to turn learners into contributors. If the guide helped you, come build it with us 👇
github.com/learntocloud
Learn to Cloud started in June 2021 as a free, open source guide to help people break into cloud. This year, we built out a custom platform on Azure that lets us offer progress tracking, more hands-on labs, and more room to grow.
A lot of this work is possible because I work at GitHub and have access to GitHub Copilot as a perk. I choose to use that privilege to invest in open source education. A few numbers that are really cool to see:
1,100+ GitHub stars on the app
~15,000 registered learners
1,000+ pull requests across the Learn to Cloud org
Some fun behind-the-scenes numbers from the app repo alone:
1,014 commits
610 PRs opened, 466 merged
437K lines added, 363K deleted (yes, we rewrote a lot)
382 commits co-authored with GitHub Copilot, more than 1 in 3
18 PRs opened by the Copilot cloud agent
178 Dependabot PRs to update dependencies
Learn to Cloud also gives me a place to get hands-on with cutting-edge technology and keep building new skills. One pattern I've been seeing is the shift toward agents working asynchronously in cloud environments. GitHub's Octoverse report found that Copilot cloud agent usage grew 230% year over year!
So I have two goals for the rest of the year.
First, get better at using the GitHub Copilot cloud agent. I want to understand what we need to improve in our repos so we can efficiently delegate more work asynchronously in a cloud sandbox.
Second, help more learners become contributors. The plight of every open source maintainer!
Very exciting times ahead :)
learntocloud.guide
github.com/learntocloud
Can't argue with execution. My point stands though - the compute goes where the power already exists. So the real question: who else is executing, not just on rockets, but on generation capacity?
Replying to @cyberrootkb @VraserX
Fair enough. More than ideas though, SpaceX is literally executing at a high level.
1
10
My take: what Ronaldo's career taught me is that no matter how great you are, detractors will always disregard that greatness. But 85% of Earth's payload to orbit doesn't lie. Just keep winning.
I pity anyone who seriously claims that anyone with enough money can do great things. It's always people full of spite, self-loathing, jealous, with small minds. I wish I could filter them out on my X feed. They bore me.
Inspire me, or shut up.
10
Kwasi Boateng retweeted
Check It Out >> Episode 16 in Series 7 of the @CorixPartners Cyber #Security Transformation #Podcast, hosted by @Corix_JC
"#Cybersecurity is a #Leadership Problem: Why the #management model needs to catch up" >> buff.ly/AaoOnBr
#tech #business #governance #CEO #CISO












































