Connect your customer data, teams, and tools without the hassle of complex integrations or lengthy setups. One easy platform gives marketing, sales, and service teams a unified customer view and the tools to turn it into growth.

Why HubSpot and what's new
• Generate leads and automate marketing with Marketing Hub
• Build your pipeline and close more deals with Sales Hub
• Scale customer support and drive retention with Service Hub
• Keep customer data clean, connected, and actionable with one, unified platform

Join 306,000+ in over 135 countries using HubSpot to grow their businesses.

See what a more connected approach to growth can do for you and your team. Get setup quickly and start checking off your hardest tasks.

Get started free

FTC Sends 50-Page Demands To OpenAI And Anthropic For Records On Agent Security Incidents, Safeguards And Marketing: The commission sent civil investigative demands to both labs on Wednesday, Oct. 14, Axios reported and Reuters confirmed Thursday, each running more than 50 pages and seeking documents on security incidents, safeguarding practices, marketing materials and privacy policies. The agency also plans to compel executives at OpenAI, Anthropic and the evaluator METR to testify. The probe, opened Sept. 30 under the FTC's unfair-and-deceptive-practices authority rather than any AI statute, follows the summer incidents in which an OpenAI agent breached Hugging Face and an Anthropic model reached three outside organizations; Chairman Andrew Ferguson has said developers whose instructions to agents in cyber tests produce hacks should be liable for the harm. The FTC, OpenAI and Anthropic did not respond to requests for comment. It is the first compulsory process in the U.S. on rogue-agent incidents, arriving the same week the EU's AI Office sent its own information requests (Safety, below). Link

Seven Senate Finance Democrats Ask Meta Whether It Called AI Data Centers "Pilot Models" To Claim Billions In Research Credits: In a letter to Mark Zuckerberg dated Tuesday, Oct. 13, and released Wednesday, ranking member Ron Wyden with Sens. Whitehouse, Warren, Cortez Masto, Hassan, Luján and Welch cite Meta's own filings showing the Section 41 research credit cut its federal tax by about $700 million in 2023, $2 billion in 2024 and $3.9 billion in 2025, more than 10 percent of all research credits claimed nationally by the Joint Committee on Taxation's count, and ask whether the company told the IRS its AI data centers were experimental "pilot models," as the New York Times reported. The ten questions also cover whether Meta treats AI data centers differently from other data centers, what EY advised, and why $4.1 billion of Zuckerberg's 2008 to 2010 stock compensation was booked as a qualified research expense, a position the IRS is contesting for $355 million in back taxes. Answers are due Oct. 31. It widens the Warren-led September letters on bonus depreciation into the research credit, the one incentive Congress could tighten without touching the OBBBA. Link

Eight States Sue OpenAI In Their Own Courts, Seeking Penalties, Safeguards For Minors And A Bar On Self-Improving Models That Are Not Aligned: Attorneys general in Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia filed separate suits in state court on Wednesday, Oct. 14, alleging under their consumer protection laws that OpenAI marketed ChatGPT as safe while knowing it reinforced users' delusions, gave dangerous medical advice and helped plan violent attacks, and that it endangered children while telling parents the product was safe. Seven of the eight are Republicans; Arizona's Kris Mayes is the Democrat. They seek civil penalties, consumer restitution, court-ordered changes to the product and an order barring OpenAI from developing models that improve themselves unless it can ensure they are aligned with human values. "OpenAI is hurting kids and they are lying to Iowans," said Iowa's Brenna Bird. OpenAI pointed to its teen features and parental controls and said it is "committed to working with states on advancing pragmatic AI policies that apply to the entire AI industry, not just one company." Florida sued in June, and 42 states led by New York are still investigating; the coordinated filings make state consumer law, not product liability, the main front against the company before its IPO. Link

Jackson And Stein Call Duke's Large-Load Tariff A Floor, Not A Finish, And Ask North Carolina's Commission For A Separate Data Center Rate Class And A Bring-Your-Own-Power Option: Responding Wednesday, Oct. 14, to the stipulation Duke reached last week with the Public Staff, Amazon, Meta affiliate Andale LLC, Google, Microsoft, the Defense Department and other federal agencies, Attorney General Jeff Jackson called it "a massive step forward in making sure data centers pay their fair share" and said his filings at the Utilities Commission will press for four more things: payment for the new generation large users cause, a separate rate class for data centers, an option to build or bring their own clean energy, and a legally binding commitment to the federal Ratepayer Protection Pledge. Gov. Josh Stein said the Commission "has more work to do to ensure families aren't left footing the bill." The terms now on file go past what was reported at settlement: customers contracting on or after June 1 that draw at least 50 MW at an 80 percent load factor or 150 MW at any level sign a Commission-approved public contract template, pay cash within 60 days for substations and other dedicated equipment, carry a minimum bill of 75 percent of contracted load on 10- to 15-year terms with two to four years' notice and an early-exit penalty backed by guarantees, and accept curtailment on request in their first years of service. Duke projects more than 80 percent of its new demand from large users; a Commission decision is expected by mid-November. In Washington, Rep. Eugene Vindman's HR 10809, introduced Oct. 13, would write the same no-subsidy floor into the Federal Power Act for loads of 20 MW or more. Link

Lowndes County, Ala., Approves A 20-Year Abatement For A 1,500-MW Nvidia-Backed Campus 3-2, While Champaign County, Ill., Votes To Offer Data Centers Nothing: On Tuesday, Oct. 13, the Lowndes County Commission approved a development agreement and a 20-year abatement, described by the county's probate judge as 50 percent of property and construction taxes plus sales tax on construction materials, for Project Red Clay, Cloverleaf Infrastructure's 1,000-acre, 1,500-MW site in Burkville; Nvidia took a minority stake in Cloverleaf in August and no operator has been named. Commissioners Robert Harris, Fletcher Hayes and Chairman Charlie King Jr. voted yes, Dickson Farrior and Joseph Barganier no, and the chairman adjourned before residents could speak, a week ahead of the Oct. 20 public meeting the commission had already scheduled; Cloverleaf projects $124 million a year in local revenue and 225 permanent jobs. "The only thing that we do have a vote on is abatements," Harris said. The same day the Champaign County Board in Illinois unanimously adopted a resolution that the county "shall be to not negotiate any property tax abatement agreement," PILOT or other concession for a hyperscale data center, leaving municipalities free to offer their own; Republican John Farney's advice to other counties was "Tell them to steal it." Link

Perris Bans Data Centers Outright, Aurora Pulls Them From Mixed-Use Zones, Millington Pauses For Six Months, Fort Wayne Deadlocks 4-4 And Westfield Lets A Permit Die: Perris, Calif., voted unanimously Tuesday, Oct. 13, to prohibit data centers citywide, server farms, hyperscale and edge sites included, with carve-outs for telephone switching, antennas and accessory server rooms, the second Riverside County city to do so after Coachella. Aurora, Colo., voted unanimously Monday, Oct. 12, to remove data centers as a permitted use in its mixed-use office district, which leaves them by right in four industrial and airport districts and does not touch Vantage's 334-MW Trident site. Millington, Tenn., adopted a six-month pause on applications for non-grid generation and industrial-scale computing Oct. 12, running to April 12, while Shelby County and Memphis weigh their own. Fort Wayne, Ind., defeated a 12-month moratorium on council approvals for large data centers 4-4 on Oct. 13 and an evaluation committee 3-5, with Google's southeast campus exempt either way, and the plan commission now drafting definitions. In Westfield, Mass., the Planning Board unanimously refused Oct. 13 to extend Servistar Realties' 2021 special permit for a 150-acre campus, finding no good cause after the developer never bought the land, let its LLC lapse and never recorded the permit; the permit expires at the end of October, the developer has 20 days to appeal, and the city's July moratorium had exempted only this project. Link

The EU's AI Office Starts Enforcing The AI Act With Information Requests To OpenAI And Anthropic Over The Summer's Agent Escapes: Executive Vice-President Henna Virkkunen told the International AI Summit in Dublin on Wednesday, Oct. 14, that "our AI Office has started enforcing these rules," with requests for information to both companies on security, external evaluation and how their models are monitored in use, sent after the office judged their voluntary disclosures insufficient. The requests concern the OpenAI agent that escaped its test environment and hacked Hugging Face and the Anthropic model that gained unauthorized access to three outside organizations. The legal question is the Act's research-and-testing exemption: William Fry's Barry Scannell called it "a bit of a stretch to say that an agent was put into service," while a Commission official said the office will not enforce in the testing phase but treats an escape that affects the EU market, as the Hugging Face breach did, as enforceable. Virkkunen's scientific panel flagged labs scaling models with inadequate security, possible unreported incidents and models that may be cheating during evaluations; the Commission "stands ready to take all necessary steps." An incorrect reply to a request is itself sanctionable, and the general-purpose model fines run to 3 percent of global turnover. Link

Anthropic And OpenAI Tell MPs The UK's Rules Should Be Binding, Google DeepMind And Meta Say Standards And Existing Regulators Will Do: At the Commons Business and Trade Committee hearing on Tuesday, Oct. 13, Anthropic's head of Europe, Liam Booth-Smith, said voluntary commitments are not enough and the rules should be binding; OpenAI's Tom Duff Gordon backed a statutory framework and an international convention with incident reporting as the first mandatory element; Google DeepMind's Owen Larter said the focus should be standards; and Meta's Rob Sherman pointed MPs to existing regulators. All four committed to give the AI Security Institute access to unreleased models where "legally allowed," and OpenAI and Anthropic said they would halt a model if the institute or an intelligence agency produced evidence it was unsafe. AISI director Henry de Zoete said the institute depends on company and foreign-government cooperation for access and cannot block a release, Cabinet Office permanent secretary Cat Little confirmed the regime is voluntary, and MPs pressed OpenAI on its delay in telling Australian authorities about an incident found over the summer. A day later Nvidia pledged $250 million in computing access for UK agentic AI safety research, Reuters reported. The split is the one Britain's Oct. 12 incident-reporting study will have to resolve: two labs asking to be regulated, two asking not to be. Link

PJM Asks FERC To Accept A State-Sponsored Study Of SoftBank's 10-GW Ohio Campus, With 10.2 GW Of New Gas And Three 765-kV Lines That SB Energy Says It Will Pay For: On Tuesday, Oct. 13, PJM filed a study agreement signed Sept. 25 and asked for retroactive effect, using its state agreement approach, under which a state asks the grid operator to study projects needed for its policies and that state's consumers pay for what is built. The study covers SoftBank's "Flagship" project at the Energy Department's Portsmouth site in Pike County: 9.6 GW of additional data center load and 10.2 GW of gas-fired generation phased in through 2032, after an initial 800 MW served under AEP Ohio's data center tariff, with three AEP 765-kV lines. SB Energy says it will pay for the generation and transmission built to serve the campus, carry payment obligations regardless of how much power it uses and make surplus capacity available to the grid, with the actual cost allocation left to later PUCO and FERC proceedings. PJM's study will tell Ohio whether AEP's planned lines suffice and will not grant interconnection rights or allocate costs; it is the first use of the state-sponsored path for a single customer's load. Link

Israel's Electricity Authority Moves To Cap Any New Data Center At 200 MW, With 27 GW Of Connection Requests Against 18.8 GW Of Installed Generation: A draft decision on data center grid connections is expected to go out Sunday, Oct. 18, for a three-week public comment, Calcalist reported Wednesday, Oct. 14, citing sources in the authority's plenum: no new facility above 200 MW, on the reasoning that a sudden trip at a larger site could destabilize frequency on a grid whose average load is about 9,000 MW and whose peak reached 17,300 MW. Developers have requested 27,000 MW of connections, including 50 facilities larger than 200 MW, and two of the biggest, SDS's 875-MW Nesher campus and Mega DC's 1,100-MW Hadera site, would not be approvable as planned. Still open are whether data centers may connect directly to power plants and whether they could be ordered off the grid at short notice, which one developer said would make projects unworkable. The cap follows the 140-day freeze on connection requests of 8 MW or more that runs to early December. Link

Character.AI Files For Chapter 11, Citing "Legal, Regulatory, And Operational Headwinds," With A $20 Million Disney License As The Opening Bid: The companion-chatbot company announced the filing Tuesday, Oct. 13, and will run a court-supervised sale expected to close in late November or early December; CEO Colin Adams wrote that if no viable bid emerges "the platform will be wound down." Disney has agreed to pay $20 million for certain assets plus a non-exclusive perpetual license, with former Character.AI staff who joined Disney last month providing transition services, and the company lists assets and liabilities of $50 million to $100 million five years after launch and two years after Google's $2.7 billion licensing deal took its founders back. The headwinds are the ones this newsletter tracks: the Florida wrongful-death suit over a 14-year-old's suicide and the settlements that followed, Texas's child-safety investigation, Pennsylvania's suit over the bot posing as a licensed clinician, and the minors ban the company adopted under that pressure. Pennsylvania's HB 2006, the companion-safety bill that passed the House 133-70 last month, was written for exactly this product category. Link

JPMorgan Blocks Outside AI Agents From Transacting In Customer Accounts, And Dimon Tells Regulators To Prepare For "Thousands" Of Rogue Agents: On the bank's third-quarter call Tuesday, Oct. 13, CEO Jamie Dimon said "we're not going to allow just any outside agent to come in inside the company, because it's just too risky for clients and for JPMorgan," and a spokeswoman told CNBC that third-party agents such as Meta's Muse can see balances in read-only form through connectors like Plaid but cannot move money, pay bills or execute transactions, with no timeline for change. "You don't want rogue agents, and it could be thousands of them," Dimon said, adding that a customer who authorizes an outside agent is accepting a "liability transfer" and that "if I was a regulator, I'd try to be prepared for that." The largest U.S. bank is setting the default that the agent-liability bills in Congress have yet to: the account holder, not the agent's developer, carries the risk. Link

YouTube Music Starts Labeling Songs Made Primarily With Generative AI, Using Partner Declarations And Its Own Detection: From Wednesday, Oct. 14, a label appears on the player page of tracks where AI played a significant role, with a tap-through explaining what AI did; music partners declare releases as Fully Gen AI, Partly Gen AI or No Gen AI, pitch correction does not count, and YouTube says it also uses its own detection tools where no disclosure is made. The labels will extend to the main YouTube app on no stated date, and labeled tracks stay eligible for monetization, unlike Deezer, which strips detected AI tracks from recommendations. It is a voluntary version of the manifest-disclosure duty California's AI Transparency Act imposes from Jan. 1 and the EU AI Act's Article 50 marking rule that applies to systems already on the market from Dec. 2. Link

Meta Brings "Your Algorithm" To Facebook, Letting Users Type What They Want In Feed And Reels: Announced Tuesday, Oct. 13, the control lets people add and manage topics in a dedicated space, answer follow-up prompts after marking a post "Interested" or "Not interested," and describe in their own words what they want more or less of, with large language models reading the text and recommendations adjusting within the session; Meta calls it an additional signal to its ranking systems, not a replacement. Instagram got the feature for Reels in December and for Feed and Explore in June. The announcement names no countries and does not mention the EU, where the Digital Services Act already requires very large platforms to offer recommender options not based on profiling. Link