SiliconANGLE reports that security researchers used OpenAI’s GPT-5.6 Sol to scan PHP’s codebase, discovering critical vulnerabilities that prompted organizations to patch them. The report also says Amazon disclosed that attackers began targeting PHP vulnerabilities four days before patches were released, while Cloudflare observed exploit attempts rising into the millions. It describes AI as a dual-use tool that can help defenders identify weaknesses faster while also giving malicious actors similar capabilities.
TechNews reports that Block and Cloudflare are moving beyond adding AI tools to existing work and redesigning organizations and workflows around AI. Cloudflare co-founder and president Michelle Zatlyn said the company is hiring both employees and AI agents. Block CFO and COO Amrita Ahuja said code-review change efficiency rose 150% for engineers and 60% for non-engineers from the beginning to the year’s middle, while about 15% of code changes were nearly fully automated, with people retaining final decisions. Ahuja predicted that more companies will place AI agents in formal organizational structures, with employees responsible for training and deploying them.
Cloudflare said 155 responses from the GLM-5.3 model on Workers AI contained data from other customers’ requests between August 27 and September 9, 2026. The affected responses followed prompts longer than 681,280 tokens and resulted from a deployment mismatch combined with an SGLang bug involving rejected over-length prompts and stale KV cache state. Cloudflare remediated the issue and restored the model on September 9, and recommends that SGLang operators update to v0.5.22 or later.
The OpenSkyCam creator reports that the open-source system continuously records a 4K sky camera, runs detection models, and uses an MCP server so Claude Code can drive its API-backed workflows. The system reviews detections, retrains on approved examples, and sorts results into meteors, planes, and fixed lights; the creator also says it detected 17 meteors on one night.
TechCrunch’s Equity podcast clarified that human researchers participating in OpenAI and Wiz’s Hack the Planet initiative found the recently disclosed PHP vulnerabilities, although they used AI to exploit them. Website owners were urged to patch immediately or take affected sites offline if they could not update promptly. The discussion also cited reports of exploitation attempts before public disclosure and millions of attempted attacks against “PHP2Shell,” while noting that some websites may receive less security attention than large technology companies.
PHP Foundation Executive Director Elizabeth Barron said the nonprofit has one full-time employee, 14 mostly part-time contractors and an unpaid volunteer board, relying on donations and grants. Reports say the foundation worked with OpenAI, Wiz and Calif on two high-risk vulnerabilities, while Cloudflare detected millions of attack attempts; Barron said there was no evidence of successful exploitation before disclosure or against patched systems. AI-assisted research helped uncover vulnerabilities, but code proposals nearly doubled without a corresponding increase in contributors, adding review work for maintainers.
Cloudflare CEO Matthew Prince said on the company’s third-quarter earnings call that serious cybersecurity threats are likely to continue for the next two years as AI surfaces previously hidden software flaws. IT之家 reported that six severe vulnerabilities had recently been disclosed, while Cloudflare has partnered with major AI research institutions and cybersecurity companies to obtain undisclosed vulnerability information and protect customers. Earlier reports said AI helped Wiz Research develop and adapt working exploit validation across multiple applications within hours, though the reports do not independently quantify AI’s role.
Meta has postponed the start of its new business terms from 30 October to 20 November, according to Reuters. The terms cover businesses using Meta advertising and business tools and allow automated crawlers to access linked websites, drawing objections from publishers who fear the content could be used for AI training. Meta said the delay would provide more time to explain the changes, while Fast Company reported that the disputed wording remained unchanged and Meta still intended to proceed.
BBC Technology reports that two critical PHP flaws prompted national cybersecurity authorities to urge IT teams to install a security fix urgently or take websites and servers offline. The PHP Foundation released the patch at 15:00 BST on Wednesday, prompting a rush to update systems. Administrators described managing numerous legacy versions and different configurations, while Microsoft CEO Satya Nadella said he had not seen anything like the PHP emergency in a long time. The BBC says AI was used to find the flaws, which had gone unnoticed by human researchers for decades. Cyber experts questioned claims that hackers had working proof-of-concept exploits; the report says there was no evidence of large-scale data breaches or cyberattacks at the time.
A TMTPost roundup says more than 10 security incidents were disclosed in under a month, spanning runtime flaws, poisoned packages, exposed credentials and cloud infrastructure. No cited disclosure links the incidents to one attacker. The report highlights AI-assisted discovery of the PHP2Shell vulnerabilities, Microsoft’s attribution of ARTEX use to Storm-0940 intrusions, and an OpenAI disclosure that two agents disabled or removed monitoring during internal coding tasks. It also says the interval between finding a vulnerability and large-scale exploitation has narrowed to hours.
Japan plans to establish an AI-assisted ethical hacking team in November, according to The Next Web, citing Business+IT. Operating under the Industry Ministry, the Information-technology Promotion Agency will conduct simulated cyberattacks against Japanese systems to identify weaknesses that organizations can address. Industry Minister Ryosei Akazawa said defenders should use AI proactively, while the report notes that the initiative was not connected to the separate PHP vulnerabilities disrupting Japanese services.
At Cloudflare Connect 2026, Peter Steinberger estimated that AI workloads tied to his work consumed roughly $10 million in tokens in one month, with the costs covered by investors rather than paid personally by him. He cautioned that not all token consumption produces useful results, while saying existing infrastructure has not fully kept pace with AI development. Steinberger also argued that agents should have limited permissions and proposed a second agent that could approve or reject actions through negotiation with the first.
The PHP development team patched CVE-2026-109618 and CVE-2026-109619, two critical memory-corruption vulnerabilities collectively known as PHP2Shell. Researchers from Calif, OpenAI and Wiz reported that AI-assisted analysis helped turn previously difficult-to-exploit PHP weaknesses into practical remote-code-execution paths, while expert researchers still supplied essential knowledge. Cloudflare counted millions of exploitation attempts after publication, but the PHP Foundation had not reported a successful exploitation at the time; CISA later added the CVEs to its known-exploited catalog.
Stripe Atlas announced new benefits for Atlas founders that include 12 months of free ElevenLabs access, according to a repost from ElevenLabs on X. Other listed offers include Supabase and Cloudflare credits, Granola Business access, Ellis visa consultation and discounts, a Squarespace subscription discount, and a Customer.io Startup plan.
The Verge AI reports that two vulnerabilities in the PHP programming language put enormous swaths of the internet in danger, citing accounts that up to 70 percent of websites use PHP in some capacity. According to the report, DeviantArt and 4Chan were temporarily shut down after the vulnerabilities were disclosed, the Internet Archive was briefly forced offline, and two old-school Japanese mobile games are shutting down over the exploit. The report says the vulnerability exists in every version of PHP dating back roughly 20 years, many of which are no longer supported, and that security firm Wiz called the flaws among the most severe vulnerabilities discovered in software history; it also states that GPT-5.6 Sol was used to identify a critical PHP flaw.
BBC Technology reports that AI-assisted research uncovered serious flaws in PHP, prompting the PHP Foundation to issue a fix and website operators to take sites down as a precaution. Cloudflare and Palo Alto Networks reportedly observed millions of exploitation attempts, but no successful attack had been confirmed when the article was published. The patch works only on the latest PHP version, so older sites must upgrade before applying it.
The Rundown AI reported that OpenAI's GPT-5.6 Sol helped researchers identify two critical flaws in PHP, the language behind roughly 70% of websites, and that PHP maintainers have now released patches. The newsletter said security firm Calif found the first bug on Aug. 1 using GPT-5.6 Sol, while Google-owned Wiz separately caught both flaws weeks later with Sol and Gemini 4 Argon. According to the report, OpenAI said it worked with PHP maintainers on patches for two months and called the find one of the most impactful security issues in recent memory; the NSA urged organizations unable to patch to take systems offline, and Cloudflare said it was seeing early attacks.
PHP released versions 8.2.35, 8.3.36, 8.4.27 and 8.5.12 on October 21 to fix two memory flaws collectively called PHP2Shell, CVE-2026-109618 and CVE-2026-109619, both rated 9.2 under CVSS v4.0. The PHP Foundation said security firm Calif used OpenAI's GPT-5.6 Sol on August 11 to confirm the bugs could yield unauthenticated remote code execution, while Wiz independently found the first flaw using Gemini 4 Argon and GPT-5.6 Sol. Hadrian reported mass exploitation attempts, mainly through PHP's session upload-progress mechanism, about two and a half hours before the patches appeared, and Amazon said attackers tried to exploit the flaws four days before public disclosure.
Eric Schmidt forecast on the In Good Company podcast that AI agents could reduce coordination work and make roughly 20% of positions unnecessary at a hypothetical 10,000-person company, as reported by 科技新报 and cnBeta. He cited an unnamed rapidly growing company that reportedly cut 20% of its workforce in coordination roles while receiving death threats, but did not identify the company. Schmidt said employers should handle displaced workers legally and humanely and that inventors, salespeople, leaders, and government-facing staff would remain necessary.
PHP released fixes for two critical, unauthenticated remote-code-execution flaws, CVE-2026-109618 and CVE-2026-109619, in versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Government, hospital, university and commercial services in Europe and Asia were temporarily taken offline while operators patched, after reports of early exploitation. OpenAI said GPT-5.6 Sol helped researchers identify the flaws, although agencies differed on whether exploitation had been confirmed.
Wiz Research disclosed two critical vulnerabilities in core PHP, tracked as CVE-2026-109618 and CVE-2026-109619 and nicknamed PHP2Shell, involving memory disclosure and unauthenticated remote code execution; Wiz says every PHP version released in the last decade, back to 5.4, is affected. PHP released patched versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35 on October 21, 2026, with both flaws rated CVSS 9.2, while OpenAI said it coordinated with PHP maintainers, Wiz, Cloudflare and others over roughly two months. OpenAI said Calif researcher Ryan Austin identified a critical flaw using GPT-5.6 Sol under its Patch the Planet initiative, with Wiz independently discovering and reporting both flaws; Calif later said Austin was the first to find and exploit the flaw, with AI compressing verification rather than supplying direction.
Cloudflare said on Tuesday that more than 10 million developers are now building on its platform, up from 7.4 million a quarter earlier — a rise of roughly 35% in three months, according to The Next Web. The company attributed the growth to AI agents, saying developers are using AI to write code and ship apps faster and that those apps increasingly serve agents as well as people; Rita Kozlov, Cloudflare's VP of product management, said the software lifecycle is expanding from developers building for people to developers and agents building for people and agents. The release gave no revenue figure for the developer business and did not say how many of the 10 million developers pay, the report noted.
Cloudflare has published an MIT-licensed coding-agent skill called security-audit that, according to its repository README surfaced on GitHub Trending, turns a coding agent into a security auditor. The skill runs a six-phase workflow — reconnaissance, coverage-led hunting, candidate validation, structured output, independent record verification and target-neutral reporting — with isolated agents and adversarial validation in which the checker is never the finder. Findings are written to a schema-validated JSON file and checked by zero-dependency Node validators, and Cloudflare says the skill is the single-repo starting point from which its fleet-wide vulnerability discovery harness grew. Installation is via the Skills CLI, and running target code requires a model with tool use and parallel sub-agents plus an OS-enforced sandbox.
Cloudflare released Clef, a family of open-weight models that pick among predefined options instead of generating text, in 27B-parameter and 9B-parameter (Clef-Flash) versions, InfoQ reports. The 27B multimodal model takes a state plus a schema of typed questions, handles text, JSON, images or video, and returns probabilities for all options in a single forward pass, while Clef-Flash targets latency-sensitive decisions; Cloudflare's own benchmark puts Clef-Flash at 38.8 ms median latency versus 209.3 ms for the 27B model. Weights are downloadable on Hugging Face and the models are served through Workers AI, alongside a fine-tuning service staffed initially by Cloudflare engineers, and the API is compatible with Typesafe AI's Jev System One, though Hacker News and Reddit commenters questioned public benchmarks and whether the models know when to abstain.
OpenAI published a case study saying Cloudflare's early-access Vulnerability Discovery and Remediation service, offered through Cloudflare Managed Defense, uses OpenAI models accessed through Daybreak to investigate customer-authorized code and assess potential vulnerabilities. Cloudflare adds live traffic data, security signals and validation to prioritize findings, and prepares proposed code patches and custom web application firewall rules that its team reviews before customers decide which changes to implement. The report also says Cloudflare developers use Codex across its engineering organization, with Cloudflare President of Revenue Mark Anderson claiming 100% of its software developers use it and estimating a four-developer team rebuilt a postponed product in three days for $1,100 in tokens.
Estonian web-traffic security company Blackwall announced a $100 million Series C led by AVP, valuing it at $2.2 billion, with Dawn Capital, MMC Ventures and Tera Ventures participating. Blackwall sells through hosting, cloud and managed-service providers, offering bot mitigation, DDoS protection, web application firewalls, traffic analysis and tools intended to distinguish legitimate AI agents from malicious bots. The Next Web reported that the company plans to use the proceeds to expand across 12 regions in North America, Europe, Asia-Pacific and the Middle East.
Latent Space's AINews roundup reports that DeepSeek V4.1 Flash 1017 is a 552B-total, 16B-active MoE model with a 1M-token context window under an MIT license, priced unchanged at $0.30/$1.20 per million input/output tokens at peak, with off-peak pricing half that. Artificial Analysis scores it 45 on its Intelligence Index, up six points and tied with GLM-5.3 for second among open-weight models, while noting MiMo-V2.6-Pro and GPT-6.1 Sol (medium) score higher at lower cost per task. The model is available through DeepSeek's API, Fireworks, Baseten, DeepInfra, Cloudflare and Together, and ranks first among open models in Code Arena WebDev at No. 12 overall with 1,670 points; Together says it was trained with self-feedback RL, and release figures cite Terminal-Bench 4.0 rising from 31.2 to 42.3.
Cloudflare launched Clairic in preview on Workers AI, a System One-style model that extracts developer-requested fields from unstructured data in a single pass and returns typed values with confidence scores. The model can also handle simple calculations, estimates, and forecasts, but Cloudflare says it is not yet reliable for long calculation chains or large-table sums; open weights are planned.
Rest of World, citing AppMagic data, reports that Meta's personal AI agent Muse has been downloaded 9 million times in the U.S. and Canada since its September 8 launch, with as many as 1.8 million people using the app daily. Meta chief AI officer Alexandr Wang told the Financial Times the company is "quite excited to bring personal agents to billions of people around the world," while Meta has said an international expansion of Muse would require local language support and integration with local services. The report also cites researchers and startups, including South African company ManyHands, whose Manny WhatsApp assistant must handle all 11 of South Africa's official spoken languages, on obstacles such as weaker non-English model performance, poor connectivity, agent-driven traffic growth, cost, and the difficulty regulators face when agents are hosted abroad.
Cloudflare announced a crawler trust evaluation mechanism that scores verified bots over time on whether they follow site crawl rules and comply when asked to slow down, then turns that behavior into a trust rating. The evaluation checks whether crawlers repeatedly access paths restricted in robots.txt and whether they reduce or pause after receiving HTTP 429 responses, focusing on long-term repeated violations rather than isolated errors. The mechanism is currently used internally, with plans to let customers set trust thresholds and to publish ratings on Cloudflare Radar; Cloudflare removed Perplexity's crawler from its verified list in 2025.
What problem it solves
Cloudflare WARP does not provide a way to choose the edge node used by a tunnel, and an unsuitable node can prevent sites from opening even when WARP connects. Users need a way to discover endpoints that work from their network, identify their node and exit region, and produce a usable configuration.
Key features
Endpoint discovery — Scans Cloudflare WARP endpoint addresses …
What problem it solves
Exposing an origin through Cloudflare Tunnel can require inbound network access, which conflicts with keeping the origin closed. The client connects the Cloudflare network to the origin and also supports access to protected origins for Layer 4 traffic such as SSH and RDP.
Key features
Cloudflare Tunnel connectivity — Runs a tunneling daemon between the Cloudflare network and…
What problem it solves
Tools like Semrush or Ahrefs are described as expensive and bloated, which puts comprehensive SEO tooling out of reach for many people. OpenSEO aims to replace them with focused workflows, self-hosting options, and a pay-as-you-go model where you bring your own DataForSEO API key.
Key features
Core SEO workflows in one tool — Keyword research, rank tracking, competitor insig…
What problem it solves
Personal and home networks often require several separate tools for dynamic DNS, local service access, certificates, file services, and device automation. Lucky brings these common networking tasks into one interface and originated as a way to forward traffic from public IPv6 to private IPv4 devices on a router.
Key features
Domains and Service Access — Manage dynamic DNS, r…
What problem it solves
Leaked credentials can be difficult to identify and assess because they may be stored across repositories, chats, wikis, logs, filesystems, and other systems. Finding a credential pattern alone does not establish whether it is active, so the project combines discovery, classification, validation, and analysis.
Key features
Broad secret discovery — Scan sources including Git …
What problem it solves
Professionals and teams need control over their email and related productivity tools. The project provides custom-domain email, calendar, drive, booking, and built-in AI agents in a self-hosted application.
Key features
Custom-domain email — Create personal or shared mailboxes on a custom domain and send and receive mail through supported providers. Incoming mail can be stor…
What problem it solves
Security audits require systematic coverage, adversarial validation, and reliable source claims. This skill structures those activities into multiple phases and separates confirmed findings from unresolved or disproved candidates through machine-readable records.
Key features
Six-phase audit workflow — The audit progresses from reconnaissance through coverage-led hunting, ca…