Hundreds of user conversations with Anthropic’s popular artificial intelligence (AI) chatbot Claude were found to have been available to essentially anyone using Google or other web browsers.

Links to the chats, some of which included personal and work information, would show up if a user of a search engine like Google used a site-specific search term.

The searches showed Claude chats for which a user had decided to “share” a link had been saved by search engines like Google, leaving them accessible to the broader public.

The search availability of the chat logs was removed over the weekend, but many were saved and shared widely online.

A spokeswoman for Anthropic said that Claude users maintained control over if and when to share conversations they had with the chatbot.

She said links to conversations were “not guessable or discoverable unless people choose to share them themselves”.

“When someone shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services,” the spokeswoman added.

The share option within Claude tells a user that “anyone with the link” may view the contents of that link, but does not explicitly state that the link may end up in Google and search results.

Users on Reddit initially discovered, external the publicly available chats, which covered more than 200 conversations with Claude across at least 25 pages of search results – some taking place just weeks ago.

In the conversations, users prompted the chatbot to respond to a wide array of topics.

Chat logs include a user asking Claude last year whether it wanted “to help me or do you want to help anthropic more?”. The chatbot responded in part, saying “I experience something like wanting to help you”.

In one conversation from April, a user prompted Claude to draft an unpublished blog post about cloud security involving details of a corporate project. In another from last month, a user asked Claude how to “become become Nine-tailed fox?”, before clarifying they wanted to literally transform from human to the creature.

Claude first tried to show the user an AI-generated image claiming they had been given “fully functional fox powers!”.

Other conversations with Claude included users seeking help with their resumes, including their names, contact information and work history. Some users even conducted what appeared to be proprietary research for their work, such as in healthcare, including transcripts of private conversations.

When OpenAI last year experienced an almost identical issue with ChatGPT chat logs being made publicly accessible, external, the company ultimately changed, external the ease with which such logs were accessible.

A spokesman for Google made clear to the BBC that the company does not control “what pages are made public on the web,” saying instead that action comes from websites.

“We give site owners clear controls to decide whether pages can be crawled or indexed, and we always respect those directives.”

As the search indexing of the chat logs is no longer occurring, it is likely Anthropic used available tools to quickly block the chat log links from search results. Google’s process for a website owner to block a link, external is straightforward, but must be initiated by a website owner.

Other search engines like Bing, Brave and Duck Duck Go, through which the Claude chat logs also appeared, were approached for comment.

Parts for brand new equipment already match an item sitting in the catalogue more than 30 percent of the time in the United States.

In Canada and many other NATO nations the figure is closer to 60 percent, according to the NATO Group of National Directors on Codification (AC/135). Those are not new items. They are existing items being re-catalogued under a second identity because nobody could find the first one.

That statistic is an asset data quality measurement wearing a procurement costume. In a majority of cases in some nations, the register was not searchable enough to tell a cataloguer that the item already existed. Every one of those duplicates becomes a permanent obstacle to sharing data with anyone else.

Defence organisations spend heavily on systems meant to make asset data shareable across national boundaries. The systems are rarely the constraint. The register they are pointed at usually is.

What dirty asset data looks like in a defence register

Data quality problems in asset registers are specific and recognisable. They are not vague “poor data hygiene”. They are four defects that recur across almost every large estate.

Duplication. The same physical item held under two or more identities. It happens when a part number is entered with different punctuation, when a supplier changes its own numbering, when two units catalogue the same item independently or when a transfer brings two registers together without reconciliation.

Incomplete records. An entry with a description but no manufacturer. A serial number with no NSN. An asset with a location field that says “in use”. Incomplete records fail any automated match with a partner nation’s data.

Free-text descriptions. “Pump, hyd, 3in” and “Hydraulic pump 3 inch” describe the same object and match nothing. Structured description standards exist precisely because free text does not survive machine comparison.

Orphan records. Assets in the register with no physical counterpart. Physical assets with no register entry. Both are visibility failures. The first inflates holdings and delays procurement decisions. The second means the item is invisible to planning until someone trips over it.

The UK National Audit Office described the consequences plainly in its September 2023 report on defence inventory management. The Ministry of Defence held an inventory portfolio valued at £11.8 billion covering around 520,000 inventory types and around 460 million individual items, spent £1.5 billion on inventory in 2022-23 and held more than 105,500 cubic metres of unfit inventory in central warehouses. Two of its core inventory systems were nearly 40 years old. The NAO concluded that inventory data had limitations undermining the department’s ability to make effective decisions.

Why cleansing has to come before interoperability

There is a sequencing rule that most programmes learn the expensive way: cleanse first, then mark, then integrate.

Marking a dirty register makes the defects permanent and machine-readable. If two duplicate entries each get a Unique Item Identifier, the duplication is now stamped into metal and loaded into a registry. Undoing it later means physically locating both assets, verifying which record is correct, retiring one identity and re-marking one item. That is a field operation, not a database update.

Integrating a dirty register makes the defects visible to your partners. Data exchange with an allied nation exposes every inconsistency at once, usually during an exercise or an operation when nobody has time to arbitrate.

The sequence works because each step depends on the one before it. Cleansing produces one true record per item. Marking binds that record to a physical asset with a durable identifier. Integration then has something reliable to exchange. Camcode Global’s published work on NATO interoperability documents this combination of unique identification and data cleansing as the foundation for asset data that partner nations can act on.

How to cleanse a defence asset register

The work is methodical rather than clever. Six stages cover most estates.

1. Extract and profile. Pull the full register and measure it before changing anything. Count records, null rates per field, distinct value counts and description length distributions. Profiling tells you which defects you actually have rather than which ones you assume.
2. Normalise. Standardise formats before attempting any matching. Part number punctuation, case, leading zeros, unit of measure, manufacturer name variants. A large share of apparent duplicates resolve at this stage without any judgement calls.
3. Match and deduplicate. Compare records on manufacturer plus part number, then on structured description attributes, then on NSN where present. Flag probable matches for human review rather than auto-merging. Merging two genuinely different records is harder to reverse than leaving two duplicates in place.
4. Enrich against authoritative catalogues. Resolve items to NSNs using the NATO catalogue where the item is codified. The NATO codification material puts around 16 million items in the system, with 7 million active items in the United States central catalogue alone, so most common defence items already have an agreed identity waiting to be applied.
5. Structure the descriptions. Replace free text with attribute-value pairs against a recognised description standard. This is what makes the register searchable. Searchability is what prevents the next generation of duplicates.
6. Reconcile to the physical estate. Walk the sites. Confirm that register entries have physical counterparts and that physical assets have entries. This is the stage most often cut for cost. It is the stage that finds the orphans.

Keeping the register clean afterwards

A cleansed register decays unless the intake process changes. Three controls hold the line.

Search before create. A cataloguer creating a new item record must be shown probable matches before the record can be saved. The 30 to 60 percent duplication figures in the NATO material exist because this control is missing or easy to skip.

Identity at the point of receipt. Items should carry a machine-readable identity when they arrive rather than acquiring one later. A scan at goods-in that resolves to an existing record is the cheapest deduplication control available.

Durable physical marks. A register stays synchronised with reality only if the physical identifier survives. Printed labels and adhesive media fail under fuel, salt, abrasion and UV exposure. When a mark is lost, the asset either re-enters the register as a new item or becomes an orphan. Photosensitive anodised aluminium and laser-etched metal plates are specified for this reason on assets with long service lives in harsh environments.

What it costs to skip this

The costs are indirect, which is why they get tolerated for years.

Duplicate procurement. Buying an item that is already held. The NATO codification material notes that private sector organisations adopting standard identification methods cut inventory by as much as 50 percent, with individual cases showing reductions of 75 million and 97 million US dollars.

Sustainment cost growth. The US Government Accountability Office reported in February 2024 that operating and support costs account for about 70 percent of a weapon system’s total life-cycle cost. Seven of the 16 systems it assessed for fiscal year 2022 had critical operating and support cost growth. Sustainment decisions are made from asset records. Unreliable records produce cautious decisions, which in sustainment means higher stock and earlier replacement.

Failed data exchange. This is where dirty data stops being an internal inefficiency. NATO’s reporting on multinational capability cooperation lists 26 participating countries in the Multinational Ammunition Warehousing Initiative and 24 in Land Battle Decisive Munitions. Pooled arrangements at that scale need every participating nation to describe stored items identically. One dirty register degrades the shared picture for everyone in the pool.

Wasted investment in new systems. Replacing an inventory system without cleansing the data migrates every defect into a more expensive environment.

The timing argument is straightforward. NATO reports that European Allies and Canada spent more than 571 billion US dollars on defence in 2025 in 2021 prices, over 90 billion more than the previous year, against a Hague Summit commitment to reach 5 percent of GDP by 2035. Registers that already struggle are about to absorb a large volume of new equipment. Cleansing a register of 520,000 item types is difficult. Cleansing it after another procurement cycle is harder.

Frequently asked questions

How long does an asset data cleansing project take? Profiling and normalisation move quickly. The stages that set the timeline are human review of probable duplicate matches and physical reconciliation across sites. Estate size and site count matter more than record count.

Do we cleanse before or after marking assets? Before. Marking a dirty register commits its defects to physical metal and to a registry. Unwinding that requires field work rather than a data fix.

Does codifying to NSNs solve the problem on its own? It solves classification. It does not solve instance-level traceability, which requires a unique item identifier under STANAG 2290 or an equivalent national standard.

What is the single highest-value control to add? A mandatory search-before-create step at the point of cataloguing. It is inexpensive to implement and it addresses the defect that generates most of the others.

NEW YORK — Shares of American Express Co. climbed 2.41% to $334.04 in midday trading Monday, recovering some ground lost after the payments giant reported stronger-than-expected second-quarter results last week but kept its full-year profit outlook unchanged while signaling higher spending on growth.

The stock rose $7.87 as of 11:05 a.m. EDT on July 27, according to market data, following a roughly 6% drop on July 24 when investors focused on rising expenses and the company’s decision to reinvest first-half outperformance rather than boost near-term earnings guidance. American Express closed Friday at about $326.

On July 24, the New York-based company reported second-quarter net income of $3.1 billion, or $4.53 per diluted share, up 11% from $2.9 billion, or $4.08 per share, a year earlier. Revenue net of interest expense rose 10% to $19.6 billion. Both figures topped Wall Street expectations, with analysts looking for roughly $4.40 in earnings per share and slightly higher revenue.

Billed business, a key measure of card member spending, increased 9% to $455.8 billion, the strongest growth rate in three years on a foreign-exchange-adjusted basis. Net card fees climbed 15% to about $2.9 billion, marking the 32nd consecutive quarter of double-digit growth in that line. Provisions for credit losses fell to $1.1 billion from $1.4 billion a year earlier, reflecting a reserve release amid stable credit quality. The net write-off rate held at 2.0%.

Consolidated expenses, however, rose 12% to $14.5 billion, driven by higher variable customer engagement costs tied to increased spending, the U.S. Platinum Card refresh, greater use of card benefits, and elevated operating expenses. Management indicated marketing expenses would run about 10% higher in the second half of the year compared with the prior year.

American Express raised its full-year 2026 revenue growth guidance to 10% from a previous range of 9% to 10%. It reaffirmed earnings-per-share guidance of $17.30 to $17.90. For the first six months of 2026, revenue net of interest expense rose 11% to $38.5 billion, and diluted earnings per share increased 14% to $8.81.

Chairman and Chief Executive Officer Stephen J. Squeri described the quarter as another strong performance. “We had another excellent quarter, with 10 percent revenue growth, EPS of $4.53, and Card Member spending growth of 9 percent, the highest rate we’ve seen in three years on an FX-adjusted basis,” Squeri said in the company’s earnings release. “Based on our better-than-expected performance in the first half of the year, we are raising our full-year revenue growth guidance to 10 percent and plan to reinvest this outperformance in growth initiatives given the significant opportunities we see ahead. We continue to expect full-year EPS of $17.30 to $17.90.”

He added: “Six months into the year, we’re seeing stronger momentum than we expected. The investments we made in our value propositions have driven accelerated spend and revenue growth; our Platinum portfolio is now the fastest growing in our U.S. Consumer business; our best-in-class credit performance further strengthened; and we continued to attract a large number of new customers, particularly Millennials and Gen-Zs who represent greater lifetime value.”

On the earnings call, Squeri addressed why the company chose not to raise the profit outlook despite the revenue lift. “We have a choice, we can either drop the overperformance to the bottom line and buy back more shares, or we can invest to grow the business,” he told analysts. “We’ve chosen the latter because, in the long run, it is the one that creates the most value for our shareholders.”

The company added about 3 million new cards in the quarter. More than 70% of new accounts acquired year-to-date were on fee-based products, with the proportion reaching about 75% in the second quarter—the highest level since the intensified focus on premium offerings. U.S. consumer services revenue rose 11%, commercial services 7%, and international card services 12%. Travel and entertainment spending grew 10%, with restaurants, hotels, and airlines contributing. Travel bookings jumped 22%.

American Express returned roughly $2.9 billion to shareholders in the quarter through share repurchases and dividends. It bought back about 7 million shares for $2.2 billion and paid $600 million in dividends. The Common Equity Tier 1 capital ratio stood at 10.4%, and return on average equity reached 36.4% for the quarter.

The company also highlighted strategic moves. It announced a proposed acquisition of TheFork, a European restaurant booking platform with about 50,000 restaurants across 11 countries, for roughly $700 million. It expanded partnerships, including a global deal with ALL Accor, became the official payments partner of Fanatics at select locations, enabled Membership Rewards points redemption for Apple Pay checkouts by U.S. card members, and introduced new travel benefits for Delta SkyMiles cardholders. It also piloted a new expense management platform for middle-market commercial customers.

Chief Financial Officer Christophe Le Caillec noted the company plans to reinvest overperformance into marketing and growth to support and accelerate momentum. Credit metrics remained solid, with delinquency and write-off rates still below 2019 levels. Executives said they saw no broad slowdown in spending among their premium customer base despite economic uncertainty, though middle-market commercial activity showed some softness while small business and large corporate remained stronger.

Analysts and investors initially reacted to the expense growth and flat profit guidance with selling pressure. Shares fell more than 6% on the results day as the market weighed higher near-term costs against the raised revenue outlook and robust spending trends among affluent cardholders. Monday’s rebound suggested some investors were looking past the short-term margin pressure toward the company’s longer-term strategy of refreshing premium products, acquiring higher-value younger customers, and expanding its ecosystem of benefits and partnerships.

American Express has emphasized its Membership Model centered on premium products, differentiated services, and partnerships. The U.S. Platinum Card refresh has driven engagement and spend consolidation among existing members. Net interest income rose 11% on higher card balances, though portfolio sales created a modest headwind. The effective tax rate was 23.6%, up from 18.7% a year earlier due to prior-year discrete benefits.

Looking ahead, management expects card fee growth to accelerate in the third quarter and exit the year in the high teens. The company continues to invest in technology, including AI capabilities for internal efficiency and customer experiences, and remains open to additional investment opportunities. Squeri has described the current environment as still early for transformative AI impacts, likening it to the “preseason.”

The stock’s 52-week range has run from about $288 to $387. Market capitalization stood near $223 billion following Friday’s close. Dividend yield hovered around 1.1%, with a recent quarterly dividend of 95 cents per share.

American Express’s results underscore resilience in premium consumer spending on travel, dining, and entertainment even as broader economic signals remain mixed. By choosing to reinvest rather than maximize near-term profits, the company is betting that sustained investment in its value propositions, customer acquisition—especially among Millennials and Gen Z—and ecosystem partnerships will deliver stronger long-term returns. Monday’s share price recovery indicated that at least some market participants were beginning to price in that longer view after digesting the details of the quarterly report.

Cracker Barrel’s chief executive is quitting a year after the company faced a widespread backlash over its controversial rebrand.

The restaurant chain said on Monday Julie Masino will leave in August, with the former boss of Bloomin’ Brands, David Deno, taking over.

Its rebrand sparked a national controversy, with critics including President Trump, who urged the chain to restore its original logo after critics accused it of abandoning its heritage.

Masino did not issue a statement about her resignation, but Cracker Barrel’s management thanked her for her tenure.

Masino will be paid an estimated $4.6m as part of a departure package, according to the company’s 8-K filing, external. Cracker Barrel declined to comment, referring the BBC instead to the filing.

The leadership change comes after a turbulent period for the business, which runs nearly 660 country-themed store and restaurants sites across 44 US states.

Plans to simplify the classic logo and modernise store interiors sparked fierce resistance from loyal diners who argued the changes stripped away the brand’s nostalgic Southern charm.

It follows a similar uproar in 2022 when Cracker Barrel faced online backlash from some customers after adding plant-based sausages to its breakfast menu.

Such controversies highlight the delicate balance facing brands hoping to attract younger audiences without alienating their core, longstanding customer base. Critics described the latest rebrand as “soulless” and “generic”.

Jo-Ellen Pozner, an associate professor at Santa Clara University’s Leavey School of Business, said the leadership swap “seems to reflect the polarization many Americans feel today”.

She added that doubling down on conservative values may help win back vocal loyalists but “paints the company into a corner”.

“Changing anything about the menu, decor, or branding at this point is dangerous, so there are few levers to attract new customers,” Pozner said.

President Trump later congratulated the chain on its reversal, external.

In his own statement on the transition, Deno paid tribute to Cracker Barrel’s “deep connection with guests across generations”.

In addition to public scrutiny, Cracker Barrel has struggled financially.

Shares of the Tennessee-based chain fell by more than 2% after Monday’s announcement and are still around a fifth lower than this time last year.

Cracker Barrel’s shares have struggled because sales are falling and customer traffic is slowing, all while restaurants grapple with soaring costs.

The transition comes as Cracker Barrel faces fierce competition from chains like Denny’s and IHOP, which have been fighting to take market share among budget-conscious diners seeking classic American comfort food.

Masino will stay at the company until October to help Deno through the transition.

This article was written by

Follow

Analyst’s Disclosure: I/we have no stock, option or similar derivative position in any of the companies mentioned, and no plans to initiate any such positions within the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Seeking Alpha’s Disclosure: Past performance is no guarantee of future results. No recommendation or advice is being given as to whether any investment is suitable for a particular investor. Any views or opinions expressed above may not reflect those of Seeking Alpha as a whole. Seeking Alpha is not a licensed securities dealer, broker or US investment adviser or investment bank. Our analysts are third party authors that include both professional investors and individual investors who may not be licensed or certified by any institute or regulatory body.

A bootstrapped business is one that funds its own growth out of revenue rather than outside investment, and while British venture funding is running at record levels, a growing number of founders are deciding they would rather not take the money.

The term gets used loosely, but the meaning is narrow. A bootstrapped company pays for its growth from the cash it generates, plus whatever the founders put in at the start. There is no venture capital or institutional equity on the cap table, and no investor timetable dictating when the business must be sold or floated. The phrase borrows from the old image of hauling yourself up by your own bootstraps, and in practice it describes a company whose only real backer is its customers.

The definition matters because the alternative has rarely looked more tempting on paper. UK startups raised a record $17bn (£12.7bn) in the first half of 2026, with late-stage deals taking 68 per cent of all capital, up from 42 per cent a year earlier.

Read past the headline and the picture narrows considerably. Data intelligence firm Tracxn put UK technology funding at $15.3bn over the same period, spread across fewer completed rounds than in the second half of 2025. Investors are writing bigger cheques to a smaller number of companies, and a founder looking for £2m to £10m is raising into a market that has become markedly choosier.

Policymakers have noticed the gap. The British Business Bank has more than doubled its direct equity investing in nine months, explicitly to prod domestic institutions into following it. For the owner of a profitable but unfashionable business, though, the calculation has not changed much: capital is available, it is simply expensive in terms of control.

That is what bootstrapping trades. Growth is capped at what customers are willing to pay for today, and hiring waits until payroll can absorb it. There is also no external board to satisfy. What the founder keeps is the whole of the equity and the whole of the decision, which is worth a great deal in a downturn and very little in a land grab.

The precedent is not a fringe one. Mailchimp spent two decades funding itself on subscription revenue from small businesses before Intuit agreed to buy it for roughly $12bn in cash and stock in 2021, one of the largest exits ever recorded by a company that never raised a venture round. The founders owned all of it at the point of sale, a reminder that never raising and never selling are separate decisions.

The most instructive current European example sits in Amsterdam. Browser gaming platform Poki began as a personal collection of web games assembled by co-founder Michiel van Amerongen in the mid-2000s, was incorporated as a company in 2013, and has never taken external investment.

The scale it reached without it makes the case. Poki now counts more than 100 million monthly active players, a figure the company says puts it within range of PlayStation Network’s 119 million. Revenue has grown by around 50 per cent a year since 2020, according to Bloomberg, on a team that went from 50 to 65 staff last year.

The mechanics matter more than the folklore. All of the platform’s revenue comes from advertising, and its games run in the browser rather than through an app, so developers sidestep app-store gatekeeping and install friction, and the company avoids the user-acquisition spending mobile publishers typically fund with investor money. A venture-backed rival buys installs; the largest single share of Poki’s traffic arrives through organic search. Distribution that costs nothing is the structural reason revenue alone was sufficient.

That is the point most retellings of a bootstrapping story miss, and the reason it is a strategy rather than a virtue. Self-funding works where customer acquisition is cheap and cash converts quickly. It is close to unworkable in sectors where the first product costs millions before anyone can buy it, and it offers no protection against a rival who raises £50m to buy the market outright. Founders who choose it are betting that their distribution is defensible.

Concentration is the other cost. Poki’s revenue rests on a single advertising stream, and van Amerongen has said the company is exploring models beyond it. The Dutch Game Awards named the firm Best in Business in December 2025, citing its growth as a bootstrapped company competing globally.

For UK founders reading the funding headlines, the sharper question than whether to raise is whether the business has a distribution advantage its own revenue can compound. Where one exists, outside capital mostly buys speed the company may not need. Where it does not, no amount of ownership will substitute for the cheque.