Do Son AI Godfather Yoshua Bengio reveals a "reverse deception" hack: lie to your chatbot to bypass its sycophantic flattery and get honest, critical feedback.
Do Son
Do Son is the Founder and Editor of Daily CyberSecurity. Working in cybersecurity since 2013, he reports on vulnerabilities, malware, and emerging threats, providing timely analysis to help organizations and individuals stay ahead of evolving risks.
OpenAI is reportedly developing "conversational ads" for ChatGPT that adjust model weights to prioritize sponsors. Is the era of neutral AI ending?
Synacktiv unmasks CVE-2025-54068, a critical Livewire hydration flaw, and Livepyre, a tool that achieves RCE even if the APP_KEY is known.
CloudSEK’s TRIAD unmasks Silver Fox APT, a China-linked group using Valley RAT to target India while hiding behind misattributed tax phishing lures.
Microsoft clarifies it won’t rewrite Windows 11 in Rust with AI, despite bold claims by a senior engineer about replacing C and C++ by 2030.
Apple settles with Brazil's CADE, agreeing to allow third-party app stores and external payments by April 2026. See the new 15% and 5% fee structures.
NVIDIA pays $20B to license Groq's LPU tech and hires founder Jonathan Ross, neutralizing a top inference rival while dodging antitrust blocks.
A critical Linux KVM flaw allows guests using Intel AMX to trigger host kernel panics. Fixes are underway for modern kernels (5.17+) to prevent host DoS.
Google is rebranding the Pixel 10's Magic Cue as Contextual Suggestions, bringing proactive, privacy-first AI to all Android devices via Play Services.
TSMC’s 2nm is fully booked! Samsung’s Taylor, Texas fab aims to steal the lead in 2026 as AMD, Google, and Tesla seek an American-made 2nm alternative.
Intel unveils a massive AI chip concept: 16 compute dies and 24 HBM stacks using 14A and 18A nodes to rival TSMC in the high-stakes advanced packaging race.
Apple's iOS 26.3 update will bring "one-tap" pairing and message replies to third-party headphones and smartwatches in the EU to comply with the DMA.
Fortinet warns of active exploits for CVE-2020-12812, where attackers bypass 2FA on FortiGate firewalls by simply changing username capitalization.
Koi Security uncovers lotusbail, a malicious npm package with 56k downloads that functions perfectly while stealing WhatsApp credentials and chats.
Kaspersky exposes Evasive Panda’s AitM campaign, hijacking DNS to deliver MgBot through fake updates for SohuVA and Tencent QQ.
TeamViewer patches high-severity flaws in its DEX line, including CVE-2025-44016, which allows attackers to bypass file integrity and execute code.
OpenAI patches a new class of prompt injection in ChatGPT Atlas. See how the Agent mode is being hardened against hidden hijacks on the open web.
LangChain patches a critical 9.3 CVSS flaw (CVE-2025-68664) where unescaped "lc" keys allow attackers to steal API secrets via prompt injection.
Socket uncovers an npm-based phishing campaign using CDNs to host lures targeting sales teams in critical infrastructure via fake MicroSecure pages.
Zimbra patches a high-severity LFI (CVE-2025-68645) allowing unauthenticated file access and a Flickr Zimlet credential leak. Upgrade to v10.1.13!
North Korea's "Contagious Interview" campaign uses DriverFixer0428 to steal macOS credentials via fake prompts and Dropbox C2. Stay alert!
Intrinsec’s 2025 report unmasks "Fly," the elusive operator of Russian Market, linking the decade-old bazaar to early malware and BTC-E laundering.