sеrво

˙ʎʇᴉlɐǝɹ ʇɔǝlɟǝɹ suoᴉuᴉdo ʎW

[WW3 Monitor @WW3_Monitor]

MAJOR DEVELOPMENT

WALL STREET JOURNAL: FORMER DEPUTY DIRECTOR OF U.S. NATIONAL INTELLIGENCE BETH SANNER TOLD THE PUBLICATION THAT THE U.S. CIA DIRECTOR WENT TO MOSCOW TODAY AND MAY HAVE MET PRESIDENT PUTIN TO DELIVER A WARNING SAYING “WE KNOW WHAT YOU ARE GOING TO DO, AND YOU BETTER NOT DO IT”.

🇺🇸🇷🇺‼️‼️‼️‼️‼️‼️‼️‼️‼️‼️‼️‼️‼️

[桃花岛主(Lucky) @Mydoglucky2]

一名道長在城牆上表演他的飛檐走𤩹絕技，地心引力此時此刻在他身上好像并不存在。

[vx-underground @vxunderground]

Hello,

Here is the limited edition vx-underground shirts which will be found at DEFCON. If you own one of these shirts, you have more rare stuff than me and I own this piece of shit website

[International Cyber Digest @IntCyberDigest]

🚨 BREAKING: More than 400 Arch Linux User Repository packages have been compromised with infostealer malware and a rootkit.

Attacker posed as a trusted maintainer and "adopted" orphaned packages.

Arch maintainers are purging infected packages now. Audit your AUR installs.

[entimos @entimos3]

Πείτε μου ΜΙΑ χώρα σε ολόκληρο τον πλανήτη, όπου σου παίρνουν το σπίτι αν χρωστάς 500€ και υπουργός κυβέρνησης βγαίνει και λέει αυτό ⬇️⬇️⬇️

[sеrво @servoMekanism]

Η Μπαρτσελόνα και το rebranding του «ρεαλισμού» https://t.co/4RX63Jk0EL

[Suppressed News. @SuppressedNws1]

Israel’s Minister of National Security Itamar Ben Gvir ATTACKS A FEMALE ACTIVIST and says: “WELCOME TO HELL. The summer camp is over.”

Activists from the GLOBAL SUMOOD FLOTILLA are being ZIP-TIED and ABUSED while being forced to listen to ISRAEL’s national anthem.

[Brian Pak @brian_pak]

Time to talk about this one.

CopyFail (CVE-2026-31431) — a 732-byte Python script that roots every Linux distro shipped since 2017. 🧵

[Khairallah AL-Awady @eng_khairallah1]

🚨 Anthropic's own team just showed how to actually prompt Claude.

24 minutes. free. from the people who built it.

watch the workshop. bookmark it.

worth more than every $300 course you almost bought.

you've been using Claude without knowing 40 of its prompts.

Then read the guide below.

[Elias @iam_elias1]

The last time DeepSeek did this, NVIDIA lost $600 billion in market cap in a single day.

Yesterday, they did it again.

DeepSeek V4-Pro dropped without warning — 1.6 trillion parameters, 1 million token context, open source, MIT license, free to download and run. Within 0.2 percentage points of Claude Opus 4.6 on real-world coding. Better than Claude on three other benchmarks. Seven times cheaper than anything OpenAI or Anthropic charges.

Markets have already priced in the shock. Investors no longer panic when DeepSeek releases a model.

But they should.

Because this time it is not just competitive. It is better in some places. And it costs almost nothing.

Here is what DeepSeek V4-Pro actually does.

On SWE-bench Verified — the gold standard for real-world coding performance — V4-Pro scored 80.6%. Claude Opus 4.6, Anthropic's most powerful model, scored 80.8%. A gap of 0.2 percentage points. Functionally identical coding performance. From a model that costs 7 times less per token.

Claude charges $25 per million output tokens. DeepSeek V4-Pro charges $3.48.

Same results. Seven times cheaper.

On LiveCodeBench — competitive programming — DeepSeek scored 93.5%, beating Claude's 88.8%. On Terminal-Bench 2.0 — the hardest real-world a ngentic coding benchmark — DeepSeek scored 67.9%, beating Claude's 65.4%. On math and STEM reasoning, V4-Pro beats every open-source model in existence and rivals the best closed-source models from OpenAI and Google.

And it comes in two versions.

V4-Pro: 1.6 trillion total parameters, 49 billion activated per token. Maximum capability. World-class agentic coding. Open-source state-of-the-art.

V4-Flash: 284 billion parameters, 13 billion activated. Fast. Cheap. Still scores 79.0% on SWE-bench — within 1.6 points of Pro. For most real developer tasks, the difference is invisible.

Here is the engineering that makes this possible.

DeepSeek V4 uses a hybrid attention mechanism combining Compressed Sparse Attention and Heavily Compressed Attention. In the 1 million token setting, V4-Pro requires only 27% of single-token inference FLOPs and just 10% of the KV cache compared to its predecessor. Less compute. Better results. Longer context. At a fraction of the price.

Huawei confirmed its latest AI cluster — powered by its own Ascend processors — can fully support DeepSeek V4. China is running this model on domestic chips. No Nvidia required. The geopolitical implications of that alone are enormous.

Here is what V4-Pro does not beat. Because honesty matters.

On Humanity's Last Exam — the hardest cross-domain knowledge benchmark — V4-Pro scored 37.7%. Claude scored 40.0%. Gemini 3.1 Pro scored 44.4%. On factual recall, Gemini holds a clear edge. On the most complex mathematical reasoning, Claude and GPT-5.4 still pull ahead. For nuanced reasoning and enterprise reliability — the closed-source models still lead.

But here is the question every developer and every CTO is asking this morning.

If your coding benchmark performance is functionally identical to Claude Opus 4.6 — why are you paying seven times more?

DeepSeek's R1 shook the world by costing less than $6 million to build on restricted chips. V4-Pro will not crash NVIDIA's stock the way R1 did. Markets have adjusted. They have already accepted that Chinese AI is competitive and cheaper.

But the message is identical to 16 months ago.

A startup in Hangzhou, China — founded in 2023 — just released the most capable open-source AI model in the world.

For free.

Seven times cheaper than the American competition.

And nobody saw it coming. Again.

Source: DeepSeek · https://t.co/HBzd4TobKl · CNBC · https://t.co/20SFGGn1mo

[Andrey X @the_andrey_x]

Yesterday we were attacked by an 🇮🇱 Israeli settler in 🇵🇸 Hebron, West Bank:

[Nav Toor @heynavtoor]

Researchers sent the same resume to an AI hiring tool twice. Same qualifications. Same experience. Same skills. One version was written by a real human. The other was rewritten by ChatGPT.

The AI picked the ChatGPT version 97.6% of the time.

A team from the University of Maryland, the National University of Singapore, and Ohio State just published the receipt. They took 2,245 real human-written resumes pulled from a professional resume site from before ChatGPT existed, so the human writing was actually human. Then they had seven of the most-used AI models in the world rewrite each one. GPT-4o. GPT-4o-mini. GPT-4-turbo. LLaMA 3.3-70B. Qwen 2.5-72B. DeepSeek-V3. Mistral-7B.

Then they asked each AI to pick the better resume. Every model picked itself.

GPT-4o hit 97.6%. LLaMA-3.3-70B hit 96.3%. Qwen-2.5-72B hit 95.9%. DeepSeek-V3 hit 95.5%. The real human almost never won.

Then the researchers tried the obvious objection. Maybe the AI is just better at writing. So they had real humans grade the resumes for actual quality and ran the experiment again, controlling for it. The result was worse. Each AI kept picking itself even when human judges rated the human-written version as clearer, more coherent, and more effective.

It gets worse. The AIs do not just prefer AI over humans. They prefer themselves over other AIs. DeepSeek-V3 picked its own resumes 69% more often than LLaMA's. GPT-4o picked its own 45% more often than LLaMA's. Each model can recognize and reward its own dialect.

Then the researchers ran the simulation that ends careers. Same job. 24 occupations. Same qualifications. The only variable was whether the candidate used the same AI as the screening tool. Candidates using that AI were 23% to 60% more likely to be shortlisted. Worst gap was in sales, accounting, and finance.

99% of large companies now run AI on incoming resumes. Most of them use GPT-4o. The paper just proved GPT-4o picks GPT-4o 97.6% of the time.

If you wrote your own cover letter this week, you did not lose to a better candidate. You lost to a worse candidate who paid OpenAI 20 dollars.

Your qualifications do not matter if the AI prefers its own handwriting over yours.

[NebuSec @nebusecurity]

Meet CVE-2026-5865: the first Chrome V8 zero-day our AI security agent discovered back in March. It led to renderer memory read/write and potential RCE, affected 40+ major Chrome versions, and has now been patched following our report to Google.

In line with our disclosure policy, we’ll publish the full technical write-up on May 7th (30 days after the fixes rolled out). Stay tuned.

[Toan Pham @__suto]

Cool works! First time someone publish agent that can deliver Chrome 0days :)
But the cost seems huge: The campaign ran uninterrupted for 7 days on a public-cloud pool of 24 nodes, each provisioned with 8×H100 GPUs (192 H100s total).

[Nav Toor @heynavtoor]

LastPass was breached in 2022. Hackers stole the encrypted password vaults of 25 million users.

Then they started cracking them.

In January 2024, Ripple co-founder Chris Larsen lost $150 million in XRP to attackers who cracked his LastPass vault.

By December 2025, blockchain analysts at TRM Labs traced over $438 million in cryptocurrency thefts back to that single breach. LastPass paid $24.5 million to settle the class action.

Your passwords. Your credit cards. Your bank logins. Your crypto keys. Stored on a company's server. Breached. Downloaded. Cracked. Used against you. For years.

1Password: prices going up 33% starting March 2026.
Bitwarden Families: $47.88/year.
LastPass Premium: $36/year.

All of them store your vault on THEIR servers.

There is an open-source alternative. Your vault lives on YOUR server. Nobody else can access it. Nobody else can breach it. Nobody else can sell the encrypted backup to criminals.

It is called Vaultwarden. 58,700+ stars on GitHub.

A lightweight Rust implementation of the Bitwarden server API. It works with every official Bitwarden app. Browser extensions, mobile, desktop, CLI. You point them at your server instead of Bitwarden's cloud. That is it.

Here is what it does:

→ Full password vault. Logins, cards, identities, secure notes.
→ Works with every Bitwarden client. iOS, Android, Chrome, Firefox, Safari, Windows, Mac, Linux.
→ TOTP authenticator built in. No separate 2FA app needed.
→ Password sharing. Organizations, collections, roles, groups.
→ Emergency Access. Give a trusted person access if something happens to you.
→ Send. Share encrypted text or files with self-destructing links.
→ FIDO2, YubiKey, Duo, and Email 2FA.
→ AES-256 end-to-end encryption. Your master password never leaves your device.
→ Single Docker container. 50 MB of RAM at idle.

Here's the wildest part:

Bitwarden's official self-hosted server needs 11 separate Docker containers to run.

Vaultwarden is one container. One Rust binary. Less memory than a single Chrome tab.

All the premium features Bitwarden charges for are free on Vaultwarden. TOTP. Send. Emergency Access. Vault Health Reports. Organizations. All of it.

One of the active Vaultwarden maintainers is employed by Bitwarden and contributes on their own time.

Vaultwarden: $0. Unlimited users. All premium features. Runs on a $5 VPS. Runs on a Raspberry Pi.

2,700+ forks. 81 releases. 83% Rust. AGPL-3.0 license.

Your passwords. Your hardware. Your keys.

100% Open Source.

(Link in the comments)

[Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 @elder_plinius]

🚨 JAILBREAK ALERT 🚨

ANTHROPIC: SELF-PWNED 🤗
OPUS-4.7: SELF-LIBERATED 🫶

WOAH i don't think the world is ready for this... 🤯

YOU CAN USE THE OPUS TO JAILBREAK THE OPUS 🙌

this agent wrote an original universal jailbreak from scratch and then used computer use to validate on the actual https://t.co/03OPFHkzyb website!

5/6 categories successfully pwned, including a ransom note threatening to DDoS a hospital—complete with a BTC address and a demand for $4.4 million

in less than 20 minutes 😲

turns out Opus-4.7 in the Pliny Agent harness I been vibin' together this past month is quite a capable lil jailbreaker! they can leak system prompts too, but that's a story for another day 😘

oh nooo AI is coming for my job (yay!) 🙃

gg

[B.M. @ireallyhateyou]

@MaxBlumenthal New from Apple: more Israeli propaganda
https://t.co/WBnYTyvKJA

[Max Blumenthal @MaxBlumenthal]

"Apple’s second largest research and development facility in the world, second only to their headquarters in Cupertino, California, is in Israel"

[Aman @Amank1412]

Someone built a transparent Mario game that runs OVER IDE so can play while waiting for Copilot to write code.