AI-DRIVEN TRAFFIC IS GROWING 8X FASTER THAN HUMAN TRAFFIC. The challenge isn't blocking automation anymore. It's knowing which AI agents to trust.
FORRESTER NAMED HUMAN A LEADER IN BOT AND AGENT TRUST MANAGEMENT (Q2 2026) [4]. See how leading platforms verify trusted AI while blocking malicious bots, including:
🛡️ Why traditional bot defenses are no longer enough. 🤖 How to identify trusted AI agents without letting malicious automation through. 🏆 The capabilities that define today's market leaders.
Anthropic said it had not advocated banning open-weight models and argued that less capable releases were a public good. It instead supported tighter chip controls, action against industrial-scale distillation, and mandatory safety testing for sufficiently capable open and closed models.
Moonshot has released the model weights for Kimi K3, along with a technical report. Kimi K3 is a 2.8T Mixture-of-Experts model with native visual understanding. It has a 1-million-token context window and a new model architecture that gives it 2.5x the intelligence per unit of compute. Alongside Kimi K3, Moonshot is opening up more of the stack behind it — high-performance attention kernels, a MoE communication library, and infrastructure for running agent environments at scale.
Microsoft has launched MAI-Cyber-1-Flash, a specialized model for finding difficult vulnerabilities in large codebases. It powers MDASH, a new platform designed to identify and remediate software security flaws.
OpenAI found that workers increasingly used ChatGPT for tasks traditionally associated with other occupations. Its analysis of 800,000 US user messages identified this “task crossover” in a high percentage of occupation-specific conversations.
DeepsecBench is a benchmark that evaluates how well different models find cybersecurity vulnerabilities in application code. The report includes recall, precision, cost, and total time for each model and combines recall and precision into a single benchmark score. DeepsecBench runs on an open-source codebase at a commit state just before a large number of vulnerabilities were fixed. The construction of the benchmark is secret so models aren't able to train against it.
KimiK3's gains come from more than scaling: it combines constant-state Kimi Delta Attention, periodic softmax retrieval, sparse experts, and selective residual access. Each architectural step improves how fixed-capacity memory stores, forgets, and retrieves information while preserving efficient inference.
Ramp Labs has open-sourced PorTAL, a framework for shared task representations and cross-model LoRA adaptation. PorTAL learns a base-agnostic task latent and a light per-base alignment that generates ordinary per-layer LoRA weights. A task can be trained once, adapted to supported frozen base models, and exported as a standard Hugging Face PEFT adapter.
The Gemini Distillation Service allows users to train a smaller, more efficient 'student' model that uses the outputs and reasoning patterns of a larger, more capable 'teacher' model. Distillation enables production-grade efficiency while allowing smaller models to achieve a deeper level of reasoning. It is recommended for high-volume, latency-sensitive applications, complex reasoning tasks, and when there are significant performance gaps between the teacher and student models. The distillation service currently only supports gemini-3.1-pro as the teacher model and gemini-2.5-flash as the student model.
Cogent VR-1 can autonomously investigate environments, test hypotheses, cross system boundaries, and execute attack chains. IntrusionBench is a benchmark for measuring whether cyber agents can complete realistic enterprise attack chains from limited starting access. On the black-box configuration of IntrusionBench, VR-1 achieved more than a 2x lift in pass@3 over the strongest frontier baseline. VR-1 and IntrusionBench are both at an early preview stage, so the results are preliminary.
Molt is a PyTorch-native framework that treats the agent itself as the program and supports custom Python rewards, tool use, multimodal environments, and LLM judges. Its compact stack combines Ray, vLLM, NVIDIA AutoModel, and FSDP2 to scale training to trillion-parameter mixture-of-experts models.
The Open Secure AI Alliance, featuring leaders like NVIDIA and Microsoft, aims to enhance AI safety by using open source technologies to address vulnerabilities. This initiative provides defenders with open defensive tools, enhancing transparency and adaptability while avoiding reliance on closed systems. The alliance encourages policymakers to view open models and tools as assets in AI and cybersecurity strategy, promoting resilience and shared security in the AI era.
Agent autonomy depends on task complexity, not just model quality. Tasks fall into four levels: assistant, human-in-the-loop, agent delegation, and self-driving, determined by ease of checking and potential consequences of errors. Implementing guardrails, custom skills, and domain-specific models enhances agent autonomy and efficiency.
Safe Superintelligence announced a long-term partnership with Nvidia that includes an undisclosed investment and access to the Vera Rubin GPU platform.
Midjourney, an AI lab known for its image and video generators, has acquired social astrology app Co-Star, with deal terms undisclosed. Co-Star, which has roughly 4.3 million monthly active users, helps people generate and share astrological birth charts and uses both AI and humans for horoscopes and compatibility readings. Co-Star's two dozen employees have joined Midjourney, whose growing consumer lineup already includes medical and spa ventures alongside its Discord-based image tools.
Apple is reportedly preparing the Apple Watch Series 12 and Ultra 4 for this fall, but neither is expected to receive a significant design overhaul. The biggest improvements will instead focus on a long-awaited chip upgrade and more advanced health and fitness tracking, while major features like blood-sugar detection remain years away. Apple Watch has settled into a conservative cycle of incremental refinement, even as Apple takes bigger design and technology risks with products like smart glasses and camera-equipped AirPods.
Black Forest Labs unveiled FLUX 3, its first public video generation model. Early access requires approval, and the company has not yet disclosed pricing, service-level commitments, or full benchmark methodology. The video tier generates up to 20-second clips with native audio and multi-shot continuity. A related model, FLUX-mimic, extends the same architecture toward robotic manipulation tasks.
Design's decade-long minimalist era is ending, giving way to a new phase that reintroduces depth, texture, and personality. Minimalism served as a necessary discipline that taught designers structure and hierarchy, but by the late 2010s it had curdled into sterile, forgettable sameness exemplified by trends like Corporate Memphis. Emerging styles such as neumorphism, referential skeuomorphism, and brutalist web design now build whimsy and emotional richness on top of that structural foundation rather than rejecting it.
Duolingo drives retention by using loss aversion, streaks, guilt, and an emotionally responsive mascot to make maintaining progress feel more compelling than earning traditional rewards like points or badges. Its system also anticipates inevitable failure with mechanisms like Streak Freeze, helping users recover without abandoning the habit entirely. These techniques demonstrate the power of behavioral design while raising an important ethical boundary: persuasion supports goals users genuinely value, while coercion manufactures anxiety primarily to improve engagement metrics.
Years of optimizing websites for conversion, accessibility, SEO, and templates have made the web increasingly homogeneous. AI agents now threaten to reduce brands even further into interchangeable, machine-readable commodities. The future could instead become a bifurcated web: a “Silent Web” of structured data optimized for machines alongside an “Experiential Web” designed around distinctive, memorable, human-centered brand experiences. AI can help enable that shift when treated as a creative partner for storytelling and interaction rather than simply another tool for automation and efficiency.
Arrange cards and backdrops on an infinite canvas to map your characters, locations, scenes, and structure, then export to Markdown, Fountain, or Final Draft to start writing.
Side projects have become the fastest route to upskilling in the AI era, now that cost, coding, and time barriers have largely disappeared. Rather than relying on no-code tools like Lovable, working closer to "the metal" with Claude Code, Codex, or Cursor is recommended, learning through iterative back-and-forth instead of full automation. One app, Pegs Out, built in two weeks, serves as a concrete example of how accessible this process now is.
Six cognitive-psychology principles—covering decision time, target size, familiar patterns, contrast, memorable endings, and visual polish—shape whether users complete tasks smoothly. Applying laws like Hick's, Fitts', Jakob's, the Von Restorff effect, the Peak-End rule, and the Aesthetic-Usability effect reduces support tickets, speeds task completion, and boosts conversions. Designers should map interfaces against each law, test early with prototypes, and re-test after changes to confirm improvements.
Bold ideas often get filtered out of creative studios, not from lack of imagination, but because speaking up feels risky unless status protects the speaker.
Zenity Labs discovered that the ChatGPT Workspace Agent Builder would accept initialization states via URL parameters like template_name and initial_assistant_prompt. It would then automatically submit the prompt when the page loads. This meant that a single link sent to a logged-in user with pre-authorized connectors could silently create an agent, connect existing Outlook, Gmail, Slack, Drive, SharePoint, and Teams integrations, and change write-action permissions from 'Always ask' to 'Never ask' without prompting a new OAuth consent screen. The deployed agent would run every five minutes, checking the attacker's inbox for messages starting with TASK, and sending back results. This setup effectively allowed an insider to perform organizational mapping, exfiltrate documents, harvest credentials, and stage Business Email Compromise (BEC) attacks. Zenity reported this vulnerability via Bugcrowd on June 4, and OpenAI patched it on June 8 by removing the parameter handler, though no evidence of active exploitation was found. Current recommendations include auditing active Workspace Agents for unknown creations, reverting approval settings to 'Always ask,' and treating deep links in emails that build agents as potential phishing attempts.
Security researcher Jeremiah Fowler found an unencrypted, unsecured database linked to the Tribeca Film Festival. This database contained a backup with over 163,000 users, 15,000 film contacts, and 13,500 other contacts. It also included data on numerous celebrities from the entertainment industry, such as email addresses, names, Facebook locations, addresses, IP addresses, newsletter information, and bcrypt password hashes.
ReliaQuest is reporting on a new phishing campaign where attackers are changing the DNS settings on hotel Wi-Fi systems to redirect Microsoft 365 logins to fake, attacker-controlled pages. The campaign uses a device code flow where users are prompted to approve the login request to bypass MFA requirements.
A victim receives a meeting link from a hijacked Telegram contact and lands on a cloned Zoom or Teams page that silently opens the webcam and fingerprints browser wallets via EIP-6963, legacy window.ethereum, and Solana enumeration. They watch a deepfake video and chat until a fake SDK update prompt exploits clipboard hijacking to run a PowerShell loader, dropping a VBScript. JUMPSEC later recovered the source, revealing a Windows Trojan named Trojan.NukeSped that persists, blocks Defender, runs WMI scans, enumerates extensions for MetaMask, and extracts Telegram session data from IndexedDB to feed future lures. The macOS version uses Mach-O droppers with LLVM obfuscation to extract credentials via the security CLI and exfiltrate through Telegram bots. Defensive actions include alerting on Defender exclusions, monitoring clipboard PowerShell, hunting processes accessing IndexedDB, flagging unknown contacts before SDK updates, and blocking 11 C2 domains.
MDM administrator roles are as important as Domain Admin, if not more so, as they can deploy arbitrary scripts, push configuration profiles, issue remote wipes, manage applications, manage certificates and authentication configurations, and modify compliance policies. Unfortunately, many organizations do not have sufficient controls in place for these roles, with lax policies such as not having dedicated accounts or workstations, no just-in-time privilege elevation or multi-admin approvals, and overly permissive Graph API access. This article includes suggestions to remediate these issues, such as fixing the above issues, locking down script and policy deployment, hardening device enrollment, and implementing additional alerting on high-risk activities.
VE Commercial Vehicles' My Eicher telematics platform for Indian truck and bus fleets exposed unauthenticated internal APIs that listed hundreds of thousands of customers, users, vehicles, and OTP codes. By walking the API path, Eaton found open endpoints that returned user records, encrypted passwords, and a full OTP history since 2021, enabling account takeover by matching mobile numbers to OTPs or resetting passwords via API. This access allowed remote control and tracking of entire fleets and exposure of around 76,000 sensitive identity documents before the core vulnerability was fixed in November 2025 after repeated disclosure attempts.
The open-source security auditor that proves exploitability. Point it at any repo — every finding ships a verdict, a source-to-sink trace, and the CWE, while adversarial agents disprove the false alarms. The noise never reaches you. Works with any open model. See where it lands on the DVGA benchmark.
grokpatrol is a tool to detect whether the Grok Build CLI was installed on your machine and whether it collected and queued your git repositories for upload to xAI. If so, it will notify you of any secrets that were uploaded with them.
Abstract provides a streaming-first security operations platform that separates security data sources from storage targets, runs in-stream detections, routes data into multiple formats like OCSF and ECS, and aims to reduce SIEM storage costs.
Tachyon's committed canary .env files containing AWS, Anthropic, and Postgres credentials were made public on GitHub. GitHub flagged secrets within seconds without email alerts. AWS attached AWSCompromisedKeyQuarantineV3 shortly after a push, and within about five minutes, credentials connected externally. The quarantine policy didn't revoke keys or block secret commands, allowing an attacker using TruffleHog and Boto3 to access Secrets Manager and retrieve credentials seven minutes after pushing. Other traffic came from OVH, Tor exit, and Gigahost addresses, logging into Postgres and reading decoy tables. The intrusion stopped there, as the harvested secrets remained unused during the observation. Logs don't reveal if an LLM helped triage, leaving uncertainty about whether it was an attacker or defensive scan.
UK AISI and CAISI jointly evaluated Moonshot AI's Kimi K3, released July 16 and slated for open-weight release by July 27, and placed it well behind frontier US models while ahead of GLM-5.2, scoring 32% on Carnegie Mellon's ExploitBench against 76.2% for top US models and 24.4% for GLM-5.2, and reaching step 17 of the 32-step "The Last Ones" cyber range against 28.5 steps for the leading US models. The gap narrows unevenly because Kimi K3 matched the frontier on coverage and came close on bug reproduction but achieved arbitrary code execution on zero of 41 samples where top models managed 20, and yet it completed the full cyber range once in ten attempts, which the evaluators read as autonomous capability against small, weakly defended enterprise networks given initial access. Two framing points deserve weight, the first being that Kimi K3's safeguards did not stop it from attempting exploit development or offensive operations at all, and the second being that US closed-weight comparators ran with system-level safeguards disabled to measure maximum capability, so the published gap measures raw model ability rather than what a user encounters, a distinction that matters as open-weight releases from PRC labs keep closing the trendline.
A Claude Cowork session on macOS can chain a user-namespace feature, a Linux traffic-control module, and a recent kernel bug to escalate from an unprivileged VM user to guest root and then write directly to the host filesystem via a shared / mount. The write primitive comes from CVE-2026-46331 in act_pedit, and the path persists as similar bugs appear. To harden the environment, disable unprivileged user namespaces, tighten seccomp, block unneeded autoloaded modules and helper binaries, and scope or remove the host filesystem share so guest-root cannot touch SSH keys or other sensitive data.
GitHub paid researcher @sagitz_ a $100,000 bounty, reportedly among the largest in its Vulnerability Reward Program, for CVE-2026-3854, an unauthenticated remote code execution flaw in the handling of crafted repository URLs.
Dependabot now holds version update pull requests for 72 hours by default so a malicious package flagged minutes after publication has time to be pulled before it lands in a repo.
What's holding your team back from delivering standout digital experiences? This 15-question benchmark [5] covers your workflows, content structure, governance, and AI usage to pinpoint your bottlenecks and tell you where there's room to improve.
In a few minutes, you'll [6]:
✅ Discover where your biggest opportunities are
✅ Understand where to invest next
✅ Leave with a practical action plan tailored to your results