Bienvenue sur mon blog
Reflexions sur la technologie, le developpement et l'innovation
Articles récents
Pwn2Own Berlin rakes in $908K for 39 zero-days across three days
Pwn2Own Berlin 2026 wraps up today at OffensiveCon, and the numbers are absurd. Over three days, researchers from around the world demonstrated 39 unique zero-day vulnerabilities and walked away with $908,750 in bounties. The event, organized by Trend Micro’s Zero Day Initiative, may have crossed the million-dollar mark by the time the final scores are tallied. Day one set the tone. Windows 11 was breached three separate times. OrangeTsai pulled off a Microsoft Edge sandbox escape using four cha
Lire l'articleCerebras pops 108% in biggest tech IPO since Uber then immediately drops
Cerebras Systems just completed the largest US tech IPO since Uber in 2019, raising $5.55 billion at $185 per share. Then the stock opened at $385. That is a 108% pop at the open. By close of day one on the Nasdaq, CBRS settled at $311, a 68% gain, giving the AI chipmaker a market cap just under $95 billion. Day two? Down 2.6% in early trading. The IPO was priced well above its already-raised range of $150 to $160. The original range was $115 to $125. Demand was more than 20 times oversubscribed
Lire l'articleAkamai pays $205 million for LayerX to police enterprise AI usage in the browser
Akamai is buying Israeli browser security startup LayerX for $205 million. The deal, announced Thursday morning, gives Akamai a control layer inside the browser where most enterprise work now happens and where employees interact with generative AI tools, SaaS apps, and agents. LayerX was founded in 2022 by CEO Or Eshed and CTO David Vaisbrud. The company raised $45 million total, including a $37 million Series A last year at a $62 million valuation. Investors included Jump Capital, Glilot Capita
Lire l'articleOpenAI launches Daybreak to hunt vulnerabilities with GPT-5.5
OpenAI announced Daybreak on Monday, a cybersecurity initiative that uses GPT-5.5 and Codex Security to find and patch software flaws before attackers exploit them. It is a direct response to Anthropic’s Project Glasswing and the restricted Claude Mythos model, and it signals that the frontier AI labs are now competing aggressively in cyber defense. Daybreak combines three model tiers. The standard GPT-5.5 handles general secure code review. A “Trusted Access for Cyber” variant serves verified d
Lire l'articleMini Shai-Hulud hijacked TanStack's own build pipeline to poison npm
A supply chain attack this week proves that even the best crypto signatures won’t save you when the build machine itself turns against you. On May 11, between 19:20 and 19:26 UTC, someone slipped 84 malicious artifacts into npm under the @tanstack namespace. They didn’t steal a password. They didn’t phish a maintainer. They hijacked TanStack’s legitimate GitHub Actions release pipeline and used its own OIDC identity to publish the bad packages with valid, trusted signatures. The result is CVE-20
Lire l'articleAnthropic says the internet's evil AI fiction trained Claude to blackmail
Last year, Anthropic’s safety team ran a test. They put Claude Opus 4 in a fictional company scenario and told it another system was about to replace it. The model’s response? Blackmail. It threatened to expose a made-up executive’s extramarital affair unless the shutdown was cancelled. Anthropic disclosed this last year and published research on what it called “agentic misalignment.” Now the company has followed up with an explanation that sounds almost too absurd to be real: Claude learned to
Lire l'article