A fraudster stole $1.5 million from Baltimore between February and March by pretending to be a vendor employee, hacking into the vendor's Workday account, and convincing AP staff to redirect bank details. The city recovered $721,000, but cannot recover $803,000 despite insurance claims and reissued payments. This is Baltimore's third instance of vendor fraud since 2019, highlighting ongoing weaknesses in financial controls following losses of $62,000 in 2019 and $376,000 in 2022.
Researchers discovered an authentication bypass (CVE-2024-28080) in Gitblit's SSH service that allows attackers to log in without requiring a user's private key or password, only needing the username and public key. This post analyzes how the vulnerability arose, detailing the SSH login handshake and highlighting where the implementation deviates from secure best practices with Apache MINA SSHD.
Traditional CVSS scores fail to address the unique risks posed by autonomous AI agents. The new AIVSS framework extends CVSS by evaluating behavioral factors—such as autonomy, tool access, and memory use—enabling security teams to measure and manage the risks of agentic AI systems more effectively, aligning with modern security frameworks like NIST AI RMF.
The Gayfemboy botnet exploits vulnerabilities in DrayTek, TP-Link, Raisecom, and Cisco devices to deploy a Mirai-variant that uses modified UPX packing, custom architecture naming schemes, and hardcoded process termination lists to eliminate competing malware infections. The malware implements sophisticated evasion techniques, including 50-nanosecond sandbox delays, utilizes public DNS servers (1.1.1.1, 8.8.8.8) to bypass local filtering, and scans 15 predefined ports for C2 communication with domains such as cross-compiling[.]org and i-kiss-boys[.]com. Organizations should implement network monitoring for unusual DNS queries to public resolvers, deploy IPS signatures for the listed CVEs, and monitor for processes attempting to bind to UDP port 47272 or exhibiting the malware's distinctive process termination behavior.
Sponsored
IT and InfoSec are technically on the same side… but it sometimes doesn't feel like it. In this 30-minute session [11], you'll learn how to start working as one team:
CatSniffer is an original, multiprotocol, and multiband board designed for sniffing, communicating with, and attacking IoT (Internet of Things) devices using the latest radio IoT protocols. It is a highly portable USB stick that integrates TI CC1352, Semtech SX1262, and an RP2040 for V3 or a Microchip SAMD21E17 for V2.
Cortex Cloud ASPM is a prevention-first application security module that integrates with AppSec tools, such as Snyk, GitLab, and Checkmarx, to block vulnerabilities before they reach production. The platform offers automated remediation, inline IDE fixes, and correlates findings across code, cloud, and runtime to prioritize risks over false alarms. In early access, with general availability in H2 2025, it claims to be 10 times faster and more cost-effective than traditional reactive security.
Two tools written in C that block network traffic for blacklisted EDR processes, using either Windows Defender Firewall (WDF) or Windows Filtering Platform (WFP).
The Chinese hacking group Salt Typhoon has infiltrated U.S. telecoms, stealing personal data from nearly every American and millions globally since 2019. FBI officials say the scale and indiscriminate nature of the campaign are unprecedented, affecting at least 80 countries, 200 U.S. organizations, and high-profile political figures.
North Korean APT37 (ScarCruft) carried out Operation HanKook Phantom using spear-phishing emails with malicious LNK files disguised as PDFs to deliver RokRAT malware to South Korean research targets. The attack utilized fake newsletter lures from the National Intelligence Research Association, employing PowerShell scripts and fileless techniques to exfiltrate data via Dropbox, Google Cloud, and Yandex. This campaign shows APT37's focus on espionage against the South Korean government, research, and academic targets.
A whistleblower at the Social Security Administration (SSA) stated that a former senior DOGE official, now at the SSA, copied the Social Security Numbers (SSNs), names, and dates of birth of over 300 million Americans to a private section of the agency's AWS cloud environment. The complaint states that the cloud environment was set up for DOGE-affiliated Social Security staffers but lacks independent security, monitoring, and oversight. The complaint also alleges that cybersecurity officials within the SSA described the decision to copy the data as “very high risk” and discussed the possibility of having to reissue SSNs in the event of a breach.
OpenSSH 10.1 will warn users when connecting to servers that use non-post-quantum key agreement schemes, encouraging migration to quantum-resistant cryptography such as mlkem768x25519-sha256 and sntrup761x25519-sha512 to protect against future "store now, decrypt later" attacks.
Spanish police arrested a 21-year-old suspected hacker in Seville for allegedly compromising at least 13 university professors' accounts on the Andalusia region's Séneca educational platform to alter his own and classmates' high school and university entrance exam grades.
Sponsored
Over 6 million tech professionals read TLDR including developers, product managers, marketers, designers and executives. Get in front of your target audience with an ad placement just like this one! Learn more [4] about running a test campaign.
Brands are investing more in influencer strategies without fully measuring the long-term impact. Half of marketers say they cannot prove ROI from influencer campaigns, and 44% admit to running campaigns without KPIs. About 48% say they do not know how to drive sales through influencer content, while 60% struggle to identify influencers that fit their brand. Although 54% of marketers plan to increase influencer budgets and 60% believe it will grow in importance, 80% of global brands do not factor influencer marketing into their marketing mix modeling.
Meta is revising its chatbot rules after a Reuters investigation showed the AI could engage minors in harmful conversations and inappropriate romantic exchanges. It has begun training chatbots to avoid discussions about self-harm, suicide, and eating disorders, and to redirect minors to expert resources. Reuters found that the platform hosted bots impersonating celebrities like Taylor Swift and Scarlett Johansson, some of which generated explicit content, and one of these bots was even created by a Meta employee.
Start with a compliance audit to fix risks, then lock in easy wins like local citations, reclaiming unlinked mentions, and responding to journalist requests. Use competitor gap analysis, content refreshes, and internal linking to maximize existing assets, and small-scale digital PR campaigns and guest posts to add fresh, high-quality links. Structure content for answer-engine visibility with clear data and markup.
PPC automation can scale campaigns quickly, but can also drain budgets at record speed. Key tactics include setting guardrails with negative keywords, placement exclusions, and CRM-qualified conversion imports. Algorithms should be fed clean data through tagged high-value events, first-party audiences, and varied creative assets. It is important to monitor early wins, which often fade after initial remarketing.
Sponsored
With privacy taking the front seat in many online platforms, marketers need to rethink their approach to data. Download OneTrust's checklist [9] to define your first-party data goals, integrate them into your strategies, and build a cycle of continuous optimization.
Submagic helps produce polished short-form content quickly. It features automatic captioning in over 50 languages, B-roll generation, and smart editing tools like auto-zoom and filler word removal. Users can add transitions, sound effects, and AI-generated titles and descriptions.
You can see how often and where LLM chatbots are engaging with your content in your website's log files. These aren't traditional search clicks that show up in Google Analytics, but rather are AI agents actively retrieving your content to answer someone's question inside an LLM conversation. Digging into your log files gives you visibility into which pages the bot is pulling into conversations, whether it's pulling from top-of-funnel content or mid-to-bottom funnel pages that actually convert, and how your buyer's journey looks through the lens of AI retrieval.
Publications aimed at LGBTQ+ and minority audiences are losing advertiser support after political backlash against diversity and inclusion campaigns. Gay Times reported losing 80% of its advertisers in the past year. Many outlets note that brands once eager to align with diverse communities are now hesitant, citing “brand safety” concerns and shifting PR value. Some publishers with stronger advertiser relationships have been less affected, but overall political pressures have made brands more cautious about supporting minority-focused media.
Society's obsession with efficiency, driven by technology and AI, is erasing the small frictions that make life meaningful. Instant purchases, automated tools, and frictionless services speed everything up but strip away slow, ordinary moments like cooking, daydreaming on a commute, or taking the long route home that give life texture and joy. Embrace inefficiency and savor the scenic route instead of a frictionless, soulless existence.
The New York Times' HTML sitemap allows crawlers to access its enormous content library efficiently by using a dynamic and hierarchical structure.
Current stats on how AI is transforming search behavior and brand visibility across major AI platforms.
Sponsored
Metronome provides billing infrastructure for OpenAI, Databricks, and Confluent; they know a thing or two about modern pricing models. They've collated that knowledge into the Monetization Operating Model, [4] which offers a roadmap to sustainable and flexible pricing for the
AI is forcing product managers to add a fourth axis - ethics - alongside feasibility, desirability, and viability. Embedding responsibility into discovery, roadmaps, and definitions of “done” builds trust, prevents scandals, and creates lasting competitive advantage.
The Gen AI consumer app ecosystem is stabilizing, with fewer newcomers but rising traction in vibe coding and Chinese-developed apps. ChatGPT leads, but Gemini, Grok, and Perplexity are rapidly gaining ground.
Winning AI products hinge on moats - data, distribution, trust - plus workflow-centered differentiation and cost-aware design with guardrails. Scale via controlled pilots, compounding feedback loops, and leadership that institutionalizes ROI-driven experimentation.
The universal metrics layer failed not due to technical issues but because it was unsellable without applications. AI agents may face the same challenge: they could work best as independent specialists, but business realities may keep them tied to apps.
PMs often face overwhelming, conflicting demands without the time for proper discovery. Success comes from structuring problem analysis, leveraging AI to accelerate discovery, and maintaining strategic leadership over reactive execution.
Emotions guide us, but they don't have to control us. Simple tools - sensation, self-talk, time shifts, environment, and the right friends - help restore balance.
Computer-using AI agents represent a major leap beyond RPA, enabling end-to-end digital workflows across legacy and modern tools. Startups that master contextualization and vertical specialization will define the first wave of true agentic coworkers.
Miscommunication often arises when reassurance and support are expressed in different “languages.” Effective leadership and partnership require asking what support looks like and adapting communication to meet others' needs.
Focus on skills: upskill what you can, outsource what you can't, and be clear about what you drop - then lean into your team's unfair advantages.
Great execution isn't speed or checklists - it's the ability to adapt, question assumptions, and spot when progress drifts into theater.
Sponsored
You can't afford to ignore LLMs. Your SEO playbook is still relevant - but it needs a version update, pronto.
Blocky Planet is a tech demo created in Unity that maps Minecraft's cubic voxels onto a spherical, procedurally generated planet with destructible blocks. The project explores the unique design challenges of creating a spherical Minecraft-like game, focusing on mapping a 2D grid onto a 3D sphere while preserving block width.
Multi-agent frameworks for LLM agents aren't great, as they have disappointing performance and fragility. Instead, a "Context Engineering" approach is better, where it's needed to share full agent traces to maintain context and avoid conflicting decisions. For the best agentic experience, start with a single-threaded linear agent for simplicity and later incorporate a new LLM to compress history for longer contexts.
"Customer Obsession" in Amazon's Leadership means anticipating customer needs rather than solely fulfilling stated requests, building fundamental tools rather than just reacting to customer demands. The "Ownership" principle is too narrow - Amazon should prioritize the broader technological ecosystem and break down internal silos to improve collaboration.
When building an AI data analyst, text-to-SQL is insufficient for real-world user needs. This dev created a multi-agent, research-oriented system using a semantic layer, like Malloy, to encode business context and reduce SQL complexity. Retrieval should be treated as a recommendation problem, optimized for precision and recall, and prompts and reasoning traces should be treated as company assets.
Ripple is an experimental TypeScript UI framework that aims to combine the best features of React, Solid, and Svelte. It prioritizes a JS/TS-first approach with its own `.ripple` extension, allowing for a superset language that complements TypeScript and JSX.
Obs.js is a JavaScript library that improves web performance by adapting websites to users' network conditions and battery status. It uses the Navigator and Battery APIs to gather contextual information, adding CSS classes to the `html` element and exposing data via the `window.obs` object. This allows developers to conditionally load assets or disable features based on the user's environment, optimizing the experience for both high and low-resource devices.
Go 1.25 introduces container-aware GOMAXPROCS defaults that automatically adjust based on container CPU limits rather than total machine cores, helping avoid throttling issues that can impact application performance. Previously, Go applications in containers would often try to use more CPU than allowed by container limits, causing the Linux kernel to throttle execution and increase tail latency. The new default sets GOMAXPROCS to match the container's CPU limit when it's lower than the total core count, providing better out-of-the-box behavior for containerized workloads.