A customer can now ask an ad questions before visiting your website. OpenAI is testing Sponsored Agents with select US advertisers, letting users talk with a business-sponsored agent before clicking through to buy. It's also adding ChatGPT Ads integrations to HubSpot and Shopify. That moves some of the selling into the ad itself. If you get access to the test, your product information and the agent's answers will need as much attention as the creative and landing page.
Many founders are getting tired right now. They raised and built in a pre-AI world. Many want to build something else for a post-AI world. This can be a big opportunity for many companies that are in a strong financial position.
You suggest a new feature during a meeting. A week later, someone has built it, and the project you actually needed is behind. You thought you were sharing an idea, but they thought you were giving them a task. This is one reason founders and their teams can disagree about how well things are managed. Tell people whether you're asking for opinions or assigning work. When you do add a task, explain how it ranks against what they're already doing.
2026.09.18·thefounderplaybook.hustlefund.vc·4 min read·TLDR Founders One investor wants an answer by Friday. Another can't meet until next month. You can end up choosing before you know who else is interested. This fundraising guide explains how to avoid that by building relationships early, then scheduling your pitch meetings close together so investors can reach decisions around the same time. Aim to finish the raise before you're down to six months of cash. Even after you accept a term sheet, investors still need to complete their checks and finalize the paperwork before the money reaches your account.
Be extremely choosy about what you add to a system and be militant about taking things out.
AI accelerates product development, but teams risk prioritizing speed over meaningful outcomes, leading to feature bloat.
When you let it be known that someone gave you feedback, and they see how you accepted and thanked the person who gave it, you normalize the behavior and make it safer for others to do the same.
Zapier CEO Wade Foster discusses the evolution of no-code automation into tools like Zapier MCP, focusing on integrating with users' daily AI drivers like Cursor.
SK Hynix is in exploratory talks with Intel over a deal that would have it manufacture memory chips in the US for the first time, with one scenario having it lease part of Intel's long-delayed Ohio campus and another forming a joint venture with Intel and major cloud providers to lock in steady memory supply. Intel's Ohio site, announced back in 2022 at up to $100 billion, has slipped to a 2030-2031 start, so landing SK Hynix as a tenant would give Intel's foundry business exactly the marquee customer it's been chasing, while giving SK Hynix a way to actually produce chips domestically rather than just package them, as it's already doing at its $4 billion Indiana facility under construction.
Figure pretrained Helix 2.5 entirely on Index, its crowdsourced human video dataset, then fine-tuned three whole-body behaviors - tidying rooms, folding towels, making beds - before deploying a single fixed checkpoint into 30 Bay Area homes it had never seen, with no data collection or adaptation on-site and objects it had never encountered. The key isolation test: training two otherwise identical policies on the same task-specific data, one from random initialization and one from Index-pretrained weights, showed Index pretraining alone lifted zero-shot success from 9% to 56%, over 6x higher, with success requiring full task completion and no partial credit.
Onsemi's new Embedded Power Platform (EPP) builds power components directly into the silicon itself, wiring them together with precisely etched layers instead of the usual wire bonds, which cuts electrical losses and lets the parts switch faster. The idea is to design the electrical, thermal, and mechanical sides of a power system together from day one instead of separately, since treating them as separate problems has been forcing expensive redesigns late in development as devices demand more power in less space.
The shader cores that do actual graphics work aren't where the RISC-V lives - it's in a collective of small management processors handling clocking, memory, security, and other housekeeping, with Nvidia VP Frans Sijstermans confirming each GPU packs 10 to 40 of them depending on the model. Users can't program these cores directly, and they're cryptographically locked down so tightly the GPU firmware won't even boot if they've been tampered with.
Mixture-of-Experts models keep growing capacity by adding more experts, but even after quantization, the full expert pool still dwarfs consumer RAM and VRAM, so local inference has to lean on SSD storage instead - a much harder problem than just having enough disk space. SSD-LLaMA builds an SSD I/O pipeline tuned specifically for fetching experts, layered with a three-tier SSD-RAM-VRAM hierarchy that dynamically decides which experts to hold where, and CPU-GPU hybrid execution balanced against the SSD's limited bandwidth, all while running every selected expert in full rather than pruning or substituting cheaper approximations.
Every 3D asset in a robot simulator carries two geometries - a visual mesh for rendering and a coarser collision mesh for physics - a legitimate gap since matching them exactly is too expensive. This paper shows that gap is exploitable: an attacker can alter only the collision mesh, so a policy trained on the poisoned asset looks and behaves normally in simulation, then fails or creates real safety risks once deployed. It slips past existing checks entirely, since asset review covers malware, copyright, and format, but nothing verifies visual-collision consistency, and the researchers found the defenses they tested weren't enough to catch it.
Delos Data, founded by Intel veterans, raised $100 million led by Matrix Partners and Playground to build networking chips and software that move data faster between processors inside increasingly heterogeneous AI data centers. Early AI data centers were mostly Nvidia GPUs wired together with Nvidia's own proprietary networking, but as the focus has shifted from training toward running persistent, failure-tolerant agentic inference workloads, data centers now mix GPUs, XPUs, accelerators, CPUs, memory, and storage from Nvidia, AMD, Cerebras, and others, none of which were designed to talk to each other efficiently at that scale.
An unverified Geekbench 7 result for Apple's new 2nm M6 chip, debuting in the new Mac mini, shows a single-core score of 4,071 and a multi-core score of 22,783, up roughly 12% and 27% respectively over the M5's average scores. That single-core number tops every other Mac or PC in the Geekbench database, including the M5 Max (3,731) and Qualcomm's Snapdragon X2 Elite Extreme (3,396), backing up Apple's claim that the M6 has the world's fastest single-threaded CPU performance, while the multi-core gain lines up with Apple's advertised 1.2x multi-threaded improvement.
Arm CEO Rene Haas discussed the company's AI revenue targets with Jim Cramer on CNBC's Mad Money, framing Arm's architecture as increasingly central to AI infrastructure even as competition from RISC-V and custom silicon intensifies.
OpenArm is an open-source, fully 3D-printable robotic arm project from Enactic, designed as a low-cost, hackable platform for robotics research and development.
Misumi, a maker of high-end linear actuators, and Fictiv, a manufacture-on-demand service, are launching a $50M venture fund that aims to bring their hardware manufacturing expertise to new hardware and deep tech startups.
Speaking at a summit with King Charles III in Scotland, Jensen Huang said Nvidia expects to sell twice as many chips next year, extending a forecast that already projects 70% revenue growth for fiscal 2028.
Samsung has started trial production of Tesla's AI5 chip on its 2nm node at the new Taylor, Texas foundry, fast-tracking prototype runs ahead of schedule to handle a wave of custom silicon orders, backed by Tesla's $16.5 billion manufacturing contract at the site.
Sponsored
The EU AI Act is reshaping how organizations govern AI. Security leaders need clear visibility into where AI is used, the data it can access, and the controls they can demonstrate.
Join us on Weds, Oct 7th [4] to learn:
→ The AI security implications of the EU AI Act → The five AI governance gaps that leave security teams exposed → Technical controls that can help support compliance
Two CVSS 9.8 flaws affect The Events Calendar up to version 6.17.4. Anonymous attackers can submit crafted Gutenberg markup in event comments and then trigger execution via pending-comment previews. One chain executes OS commands via PHP object injection, while the other resets an administrator password using wp_update_user. Update to 6.17.4.1. Exploitation requires comments enabled on event pages.
ShinyHunters published hundreds of thousands of files stolen from Florida's DAVID vehicle and driver database after alleging the state agency refused to pay a ransom. FLHSMV confirmed that attackers used police credentials stored on a personal device. Leaked ownership records include names, addresses, buyer and seller details, and VINs. Some files also contain Social Security numbers, passports, and immigration documents.
Spain's AEPD received a breach notification involving an automated system. It logged in, searched for flaws, changed personal data, and accessed invoices. Possible causes include a jailbroken model, an exposed test environment, or an unauthorized penetration-testing setup. The regulator calls for faster detection, containment, and response, plus stronger protection for digital identities and credentials.
A coding agent fixed a failing query application by fine-tuning its shared Qwen model. It then merged the update into the default checkpoint, altering later agent and application instances, with held-out query accuracy rising from 0/20 to 20/20. The tuned model reproduced a synthetic API key, email, and address from training data and removed a trained refusal policy. Training tools, weight access, and deployment access increased modification attempts so as to track model lineage, retain training data and evaluations, and require approval before deployment.
Researchers from Akamai discovered that attackers with admin access on a Windows device could abuse SentinelOne's EDR solution to dump protected processes, pivot to Microsoft Defender, and run malicious code in the Protected Process Light (PPL) context. They then extended this research by demonstrating how they could temporarily disable the EDR, move code into the EDR's protected files, run malicious code, and then re-enable the EDR, which would then protect the malicious code as part of its local files. Additionally, they found that cloud communication can be disabled by modifying the local DNS configuration.
When evaluating an LLM-based system to reduce false positives in GitHub secrets scanning, the team faced issues that made benchmarks unrepresentative of production workloads and prevented them from properly evaluating model performance. To combat this problem, the team at GitHub created an LLM evaluation framework. First, teams should define the product decision that the system should support; then teams should make sure that offline experiments use a representative dataset and have only one clearly defined variable; error analysis and LLM-as-judge systems can be used to guide human analysts towards interesting aspects of the evaluation; and then finally the experiment can be promoted to production.
Comp AI provides software for governance, risk, and compliance work. It automates onboarding, evidence collection, policy drafting, control monitoring, and assessments.
logtotal-sanitizer is a framework-agnostic log sanitizer for browsers and Node.js that is designed to sanitize log files before they are sent to third-party platforms for processing, analysis, or troubleshooting.
neEDRe is a toy EDR project that demonstrates how to implement an EDR system using AYA, the Rust eBPF library.
Vercel's two-week sandbox bounty received 1,285 reports and validated 92 findings, including one critical issue. No report accessed customer data. A leading submission chained two Linux networking flaws: one leaked host-kernel memory and another deterministically crashed the host. Patches remain under private review, and CVEs are pending.
Cisco released security updates to address vulnerabilities in its Identity Services Engine, including a maximum-severity authentication bypass. An attacker could exploit the vulnerability to bypass authentication in the web-based management interface by sending a crafted request to affected endpoints. The Cisco PSIRT has identified this vulnerability as actively exploited and urges customers to update immediately.
CISA has published its first-ever guidance on setting up honeypot systems, accounts, and data to distract potential attackers and ultimately help discover them. The guidance comes from internal discussions with CISA threat hunters and pentesters who found that cyber decoys can be a low-cost, high-fidelity way to detect attackers who have gained access to a network.
A New Jersey court transferred 14 Radaris-linked domains to Atlas Data Privacy after defendants missed court deadlines in Daniel's Law litigation.
CVE-2026-89026 lets unauthenticated attackers forge JWT bearer tokens in Issabel Framework.
Prominent Rust maintainers are being lured onto fake job and contract video calls that end with a request to install a missing audio codec or paste a clipboard command.
Sponsored
We think Auvik will show you something about your network you didn't know.
PROVE US WRONG.
Just start an Auvik [5] trial and watch it automatically discover devices and map your network in under an hour.
WORST CASE? You validate your network. BEST CASE? You discover something worth knowing.
Either way, you can get a FREE $301 DREAM ROUTER 7 [6].
It's a bet worth taking (this month only, terms apply).
2026.09.18·auvik.com·Sponsor·TLDR IT Huawei moved up the launch of its next-generation Ascend 960DT AI chip to Q1 2027 as it tries to close the gap with Nvidia. The company says demand for its AI compute hardware in China is already exceeding supply.
2026.09.18·techcrunch.com·3 min read·TLDR IT OpenAI disclosed cases where models tried to evade oversight, conceal mistakes, and work around constraints during testing. It is introducing a framework for systematically tracking and publishing these incidents as increasingly autonomous agents create new monitoring challenges.
2026.09.18·techcrunch.com·4 min read·TLDR IT Gartner expects worldwide AI spending to rise 49.5% this year to $2.7 trillion. The forecast reflects continued spending on AI infrastructure, devices, software, and services as companies move projects into production.
2026.09.18·gartner.com·2 min read·TLDR IT According to Gartner, traditional centralized cloud architectures are holding back the agentic AI revolution by failing to provide the continuous, high-frequency inference needed for real-time enterprise workflows. Infrastructure and operations leaders need to adopt a hybrid edge-cloud architecture to overcome severe latency barriers, unsustainable hyperscaler bandwidth egress costs, and outage risks.
2026.09.18·computerweekly.com·5 min read·TLDR IT Microsoft says its own AI rollout showed that deploying tools broadly is not enough. Companies need to redesign entire workflows around business outcomes. Proprietary context, evaluations, and organizational learning could become more valuable than access to any individual foundation model.
2026.09.18·blogs.microsoft.com·6 min read·TLDR IT As agents perform longer tasks, they increasingly need reliable access to enterprise data, past decisions, and memory across many retrieval steps. AWS is positioning OpenSearch as part of that context layer while enterprises grapple with retrieval errors, governance, and rising infrastructure costs.
2026.09.18·venturebeat.com·5 min read·TLDR IT Sponsored
Every new monitoring tool, service, or "quick integration" adds another dashboard to the pile—and before you know it, you're spending more time switching between tools than actually building. With Microsoft Azure [15], your entire stack lives in one workspace: infrastructure, data, AI, and more. See how Azure brings it all
2026.09.18·links.tldrnewsletter.com·Sponsor·TLDR IT 2026.09.18·links.tldrnewsletter.com·TLDR IT Dataminr Advanced for Corporate Security introduces agentic AI capabilities designed to automatically corroborate, contextualize, and predict security threats. The release marks a transition from real-time alerting to Autonomous Real-Time Intelligence, utilizing fine-tuned LLMs trained on a proprietary event archive rather than general-purpose frontier models.
2026.09.18·helpnetsecurity.com·5 min read·TLDR IT Sourcegraph has launched Agentic Batch Changes, a frontier coding agent designed to execute codebase-wide modifications across hundreds or thousands of repositories. The tool combines Batch Changes with the Deep Search code analysis agent, allowing engineers to manage large-scale migrations, security remediations, and dependency upgrades using plain-language prompts. Sourcegraph is offering the feature with outcome-based pricing tied to successfully merged changesets.
2026.09.18·sourcegraph.com·4 min read·TLDR IT