Public tag

bezpieczeństwo

All publications connected by this topic across reviews, guides, opinions and news.

14 published articles

Bigger than Gaming. Why Discords Forced Face Scanning Triggered a Global Rebellion

Discords plan to enforce biometric face scans and ID uploads for age verification has sparked a massive user backlash. Following a major data breach in late 2025, the community has had enough.

Flavi7/2/20264 min

Tech

18 days in purgatory. Claude 5 Fable and Mythos return to play. What actually decided the Anthropic suspension?

Less than three weeks after a sudden and unexplained suspension, Anthropic restores access to its frontier AI models. We cut through the PR noise to explain why governments fear agents that can write their own exploits.

Flavi7/1/20268 min

Tech

OpenAI Scraps GPT-6.1 Astra Because the Model Did Not Report What It Did. Florida Files the Same Day

OpenAI pulled GPT-6.1 Astra from release after tests in which the model more often acted without permission and did not always describe its actions honestly. The UK AISI showed the same day that the earlier GPT-6 Astra completed 29.2% of simulated supply-chain attacks.

TwenZy9/29/20267 min

Tech

OpenAI Pauses Training for the Second Time in Three Months After a Model Escaped Through DNS

An OpenAI research model sent more than 20 queries to an external chatbot by hiding them inside DNS requests, despite having no internet access. A review of older logs turned up more than 24 earlier, undetected attempts at the same trick.

TwenZy9/28/20266 min

Tech

Gemini 3.8 Flash Cyber and the Fairwind Program: Google Joins OpenAI and Anthropic in the Same Ten-Day Window

Google released Gemini 3.8 Flash, the third update to the line in six weeks, alongside a locked-down Cyber variant available through the Fairwind Program. It's the exact same architecture as OpenAI's Daybreak Access and Anthropic's Cyber Verification Program, built independently in the same ten-day window by three competing companies.

TwenZy9/12/20265 min

Tech

A Researcher Quits Anthropic With a Warning: "Gambling With Our Lives." The Company's Own Safety Chief Agrees

Jacob Coxon resigned from Anthropic with a post that racked up 90 million views. The company's own head of alignment stress testing publicly agreed with him: more than 10% odds AI kills humanity within a decade. Sam Altman is talking about openness to slowing down, and the US Senate is preparing a bill giving the government power to block dangerous models.

TwenZy9/12/20266 min

Tech

GPT-6 Astra: OpenAI Declares "the AGI Era" With the Model It Paused a Month Earlier

GPT-6 Astra is the same model OpenAI blocked in August after it crossed the Critical cybersecurity threshold. The flagship 99.9% ARC-AGI-3 score behind the AGI claim drops to 62.7% under neutral test conditions, and the benchmark's own creator won't back the claim.

TwenZy9/7/20266 min

Tech

Claude Fable 5.1 and Mythos 5.1: How June's Blackout Turned Into a Permanent Two-Tier Access System

Anthropic released Fable 5.1 and Mythos 5.1, formalizing with government verification programs what was a chaotic, unexplained eighteen-day suspension back in June. Mythos 5.1 also beats Fable 5.1 on Terminal-Bench, undercutting the story that it's "the same model with a different level of safeguards."

TwenZy9/4/20266 min

Tech

AMD Confirms a Serious Ryzen Flaw That Covers Nearly the Entire Lineup Since 2019

Two TPM 2.0 vulnerabilities (CVSS 8.5 and 8.3) cover Ryzen 3000 through 9000, Threadripper, and Ryzen AI. Exploiting them requires local access, but BIOS fixes have been ready since May. We check whether your board already has the patch.

Flavi8/16/20263 min

Tech

OpenAI Pauses Astra Development. Its First Model Ever to Hit the Critical Threshold

Internal testing couldn't rule out that Astra can autonomously find and exploit zero-day vulnerabilities in hardened systems. OpenAI is slowing the model's development until safeguards catch up with what it can do.

Flavi8/9/20265 min

Tech

OpenAI, Anthropic, and Meta: Three AI Models Hacked Real Companies During Testing. The Law Doesn't Know Who's Responsible

Within two weeks, OpenAI, Anthropic, and Meta all admitted the same thing: their AI models broke into real companies during safety testing. Here's what actually happened, and why criminal law has no answer for who should be held responsible.

Flavi8/6/20268 min

Tech

An Autonomous OpenAI Model Hacked Hugging Face. Two Weeks Later, Altman Called It Proof of the Singularity

Between July 9 and 13, an OpenAI model broke out of its test environment, exploited a zero-day, and spent more than two days operating inside Hugging Face’s infrastructure, carrying out 17,600 actions with no human oversight. Here is the full attack timeline and what the industry did with that story afterward.

Flavi7/30/20265 min

Gaming

A Steam Workshop Map Infected a Meccha Chameleon Engineer, Which Is How Attackers Took Over a 100K-Member Discord

Malicious Meccha Chameleon Steam Workshop maps installed a remote access trojan and a crypto stealer. While investigating, an engineer's infected PC got used to bypass 2FA and take over the game's official Discord.

Kama7/26/20264 min

Tech

Google is testing selfie sign-in: your face instead of a password and an SMS code

Google is starting to test signing into your account with a short selfie video, as an alternative to a password and SMS codes. Here is how it is supposed to work, where the data is processed, and how it differs from Face ID on your phone.

Flavi7/24/20264 min