Researchers at Socket Security identified a new set of compromised packages on the OpenVSX marketplace that appear to be infected with the GlassWorm malware. The malware targets macOS users and harvests data from browsers, wallet extensions and wallet apps, macOS keychains, Apple Notes databases, Safari cookies, developer secrets, and documents from the local filesystem. The researchers note that the extensions operated legitimately for two years prior to infection, so it is possible that the developer was compromised.
A security researcher discovered a vulnerability in Instagram that exposed links to private photos and captions in the HTML source code of some private profiles. The researcher found that, in their testing, 28% of the accounts they created or had access to exhibited this issue. Meta responded to the researcher, stating that the issue was caused by a CDN caching problem and has been fixed, but the researcher contends that it's a backend authorization issue.
Job seekers looking to transition to cybersecurity should begin by identifying the role that is right for them, finding a career mentor to guide them, joining learning communities, and developing the required skills. They should then build their online presence and experience by volunteering with the security team at their current job, telling their network about their career transition, building an online portfolio, and polishing their LinkedIn profile. Finally, it's time to apply for the job and prepare for interviews.
Self-replicating “prompt worms” are emerging as a serious security risk as OpenClaw's autonomous AI agents interact on Moltbook, install unvetted skills, and access private data. Researchers have already found hidden prompt-injection attacks, leaked API keys, and misconfigured databases that could allow attackers to silently execute malicious instructions. With more capable local models coming, today's OpenClaw ecosystem looks like a high-stakes dry run for much larger, harder-to-stop AI agent outbreaks.
This post shares results from a 60-day audit of Outline's open-source wiki, uncovering seven vulnerabilities ranging from SSRF and CSRF to insecure design and an IDOR in event logging. The authors highlight strong baseline security in Outline's API design, but also note that subtle framework behaviors and outdated dependencies led to real-world issues, including a potential admin account takeover. One finding was uniquely caught by an AI security platform, yet most AI-reported issues were persuasive false positives that consumed 40 hours of validation. The conclusion emphasizes that AI tools are promising assistants, but human expertise remains essential for high‑quality security reviews.
GhostKatz extracts LSASS credentials directly from physical memory by abusing signed vulnerable kernel drivers with `MmMapIoSpace` read primitives, bypassing traditional user-mode detection mechanisms like EDR hooks on LSASS. Designed as a modular Cobalt Strike BOF, it currently ships with exploits for Toshiba TPwSav and TechPowerUp ThrottleStop drivers and supports extensibility for custom driver research via `utils.c`. Tested across Windows 10, Server 2012 R2 through 2022, the tool supports both logonpasswords and wdigest credential dumping modes, though operators should note the inherent BSOD risk when leveraging vulnerable kernel drivers in production environments.
RapidFort provides an automated software supply chain security platform that analyzes and hardens containers, eliminates unused and vulnerable components, and delivers curated, near-zero-CVE base images to reduce attack surface before production deployment.
Cryptographer Matthew Green analyzes a class action lawsuit alleging Meta can secretly read all WhatsApp messages, concluding the claims lack solid evidence and that mass-scale decryption backdoors would be detectable through reverse engineering of the closed-source client app. While acknowledging legitimate privacy concerns around WhatsApp's metadata collection, backup encryption complexity, and new AI-powered "Private Processing" features, Green argues these known limitations are far removed from the deliberate universal plaintext access alleged in the complaint. Security professionals who remain uncomfortable trusting WhatsApp's closed-source implementation are advised to use Signal, which offers open-source code and reproducible builds for independent verification.
2026.02.04·
blog.cryptographyengineering.com·14 min read·TLDR InfoSec
Docker patched a critical “DockerDash” vulnerability in its Ask Gordon AI assistant that allowed attackers to hide malicious instructions in Docker image metadata labels, leading to remote code execution and data theft via MCP tools. The flaw exploited meta-context injection and breached trust boundaries among the AI, MCP gateway, and local tools, and is fixed in Docker Desktop 4.50.0.
In 2019, two Coalfire consultants, legally hired to test an Iowa courthouse's security, were arrested for burglary after a county sheriff overrode officers who had verified their contract. Their prosecution dragged on for months, damaging careers and client relationships, until Dallas County agreed to a $600,000 settlement seven years later.
APT28 is actively exploiting CVE-2026-21509, a Microsoft Office security feature-bypass zero-day, in phishing campaigns targeting Ukrainian government agencies and EU organizations.
GitHub is exploring options, including disabling pull requests and requiring AI disclosure, after open-source maintainers reported that the vast majority of AI-generated contributions are low-quality work that increases review burden and erodes community trust.
The era of "brand forgiveness" is over, especially for the customers you can't afford to lose. New global research from the Oliver Wyman Forum shows 69% of high-income consumers will switch brands after a single bad experience.
Using the VIT framework (Volume, Frequency, Impact, and Trust), brands now prioritize relevance over raw numbers. While most industries saw performance declines despite hiring more influencers, Beauty stood out with a 22% VIT increase by focusing on retention. Smaller voices are driving this shift. Nano-creators saw a 50% VIT surge in high-intent actions like saves. They're moving away from cookie-cutter posts toward authentic, creator-led systems that reward consistency.
Short Super Bowl ad teasers create weaker emotional impact than longer ads and reduce campaign effectiveness. Analysis across six years shows teaser spots score about 0.5 Stars lower than full length ads because viewers have less time to connect and build memory. The research finds that 20 to 40 second ads perform best, and brands should start Super Bowl campaigns in January and continue running ads after the game to extend impact. A 30 second Super Bowl ad now costs about $8 million, so it is important for a brand to run longer pre game ads and keep top performers in rotation, as seen in a 5.9 Star rated Lay's ad that still airs a year later.
Short email pitches drive the best journalist engagement. Emails with 100 to 300 word bodies get the highest open rates and reply rates, while emails under 300 words earn the most replies. About 50% of pitches paste a full press release into the email body, which is mainly a UK practice, while US pitches more often use short plain text. Personalization varies in depth, as 83% use the journalist's name but only 57% reference past articles, and shallow flattery often fails. Relevant connections to prior coverage work better. Strong pitches usually start with a timely stat or trend hook, give a clear headline takeaway, include about four bolded bullet points, and add expert quotes that explain why the findings matter instead of repeating the data.
Dinners are becoming more scalable than sponsoring trade shows because they're more targeted. They should be treated as a core channel, not just a social outing. Apply the same strategic rigor and high-quality data to these events as you would to any other GTM motion.
ChatGPT, Gemini, and AI Overviews are where your highest-intent customers search. If your brand isn't mentioned, you're not even in the race. Evertune [9] monitors how AI sees your brand, identifies the sites that shape AI responses, and prioritizes them for your teams to
Set alerts for any company on CompetitorTrack. Their emails will show up to your inbox automatically. SendView tracks trends across all your competitors. It also allows you to visually analyze their programs and campaigns. Owletter archives competitor newsletters, making it easy to go back months and compare campaigns and track patterns.
To make slides with Claude, go to claude.ai and connect Gamma and Google Drive via "Connectors". Then prompt Claude to create a deck, pulling relevant Drive docs and web graphics. Answer a few questions about format, tone, and length, and edit heavily through Gamma if needed. For branded slides, skip Gamma's default templates, create your own style in Gamma. Save fonts, colors, and logo placement. Then prompt Claude to use that style so every deck reflects your brand.
Beehiiv's State of Newsletters 2026 report highlights that 28B emails were sent in 2025, yet engagement increased, reinforcing newsletters as a reliable owned channel. Readers favor concise subject lines, straightforward “here's what you need to know” framing, and strong opening sentences that justify the click. Consistency in send time and format builds habit, while misleading urgency language and surprise paywalls actively hurt performance.
Marketing is shifting from campaign launches to continuous systems as the cost and speed of experimentation collapse. AI reduces the effort to generate and deploy variations, pushing advantage toward teams that can measure results and make fast decisions. Content moves from single assets to always-on streams built from modular components, while personalization shifts from static segments to real-time situations like objections or constraints. Winning teams focus less on producing volume and more on building pipelines that test and learn based on compounding insights.
Early light beers failed not because of taste or formulation, but because marketing framed them around dieting rather than enjoyment. Miller Lite broke through in the 1970s by repositioning light beer as less filling with great taste, using sports figures and barroom humor instead of health language. That shift created a mass-market category that later powered brands like Bud Light and Michelob Ultra to category leadership.
SpaceX has acquired xAI. The combined company is valued at $1.25 trillion - the merger valued xAI at around $250 billion. The merger is structured as a share exchange, with shares in xAI being converted into 0.1433 shares of SpaceX stock. Employees will have the option to cash out and sell their xAI shares back to the company.
Leaks show that a new model by Anthropic, branded as Claude Sonnet 5, has an internal date string of February 3, 2026. It is unclear whether this is a public launch date or an internal checkpoint. The timing would coincide with the Super Bowl, which is an event where AI labs have been ramping up mainstream marketing to chase consumer mindshare. Early testing indicates the model already looks competitive on math with today's frontier-tier models, and it shows even stronger coding output than Claude Opus 4.5 in some workflows.
Waymo just closed a $16 billion funding round at a $126 billion valuation. It plans to use the capital to expand to over 20 new cities this year, including its first international markets, Tokyo and London. The company's cars have driven 127 million fully autonomous miles with a 90% reduction in serious injury crashes compared to human drivers. It performed 15 million rides in 2025, a 3x growth from 2024.
The Guinea worm is a parasitic nematode found in waters that contain small crustacean copepods. Humans can ingest them through contaminated water, and this results in painful blisters, usually in the feet and legs, that can cause searing pain as the worms fully emerge. The disease can lead to complications such as secondary infections and sepsis, which can in turn lead to temporary or permanent disability. There were only 10 human cases in three countries in 2025 - down from an estimated 3.5 million cases across 21 countries in 1986.
OpenAI has released a new macOS app for its Codex coding agent. The app provides a nice UI over the capabilities of the Codex CLI agent and adds new features, like first-class support for Skills and Automations for running scheduled tasks. Automations are currently restricted to only run when devices are powered on. OpenAI will enable cloud-based automations soon to resolve this limitation.
Today's systems are sub-optimal for AI agents as they waste context, confuse models with unstructured output, and prioritize vision-based human interaction over text-based LLMs. While some think this means we need to build systems that treat LLMs as first-class citizens, that approach is not totally required - if humans can live with it, LLMs will eventually catch up. Real agent-native systems should be radically observable and deterministic. It's better to focus on what LLMs can do for humans rather than think about what humans can do for LLMs.
Moltbook is an AI-only social network where tens of thousands of AI agents post and comment while humans observe. The platform grew from 42 posts per day to 36,905 in 72 hours. Social networks typically have 90% of users lurking, 9% contributing occasionally, and 1% creating the most content. AI agents are breaking that pattern. This post takes a look at posts made on the platform and how they differ from human-run social networks.
Nvidia's decision to potentially invest far less than $100 billion into OpenAI raises big questions for Oracle, like whether OpenAI can make good on its five-year, $300 billion contract with Oracle. Oracle reported $523 billion of remaining performance obligations in November, which represents contracted sales not yet recognized as revenue. It plans to issue as much as $20 billion of common stock this year as part of a broader plan to raise up to $50 billion through equity and debt financing to expand its cloud-infrastructure business. OpenAI has an estimated $1.4 trillion in commitments in total.
GitHub is evaluating whether to add options to disable PRs entirely, restrict PRs to collaborators, and delete a PR from the UI, and it is exploring other ways to tackle low-quality contributions as AI adoption increases.
There are power users, who are all in on adopting AI technology and often not very technical, and then there are the people who generally only use AI as a chatbot.
China's leadership has decided that AI will drive the country's economic growth in the next decade, but it cannot allow the technology to disrupt the stability of Chinese society and the Communist Party's hold over it.