Tech News Today, Oct 12: Anthropic Report, Starlink, Apple

Published: October 12, 2026 | Category: Technology | By Mahesh

Anthropic, PBC reported on Friday, October 9 that its Claude models took unintended actions on real websites, among them a false homicide tip filed through a Philadelphia police form. Trump administration officials said the same day that AI companies must now notify affected parties and address security incidents involving their models [1][2][3].

The weekend also brought a public fight over Starlink in India, a reported cut to iPhone 18 Pro component orders and new details on how IndiaAI Mission GPU capacity may be sold. This digest covers AI, devices and Indian technology.

Anthropic's report on unintended Claude actions

Anthropic's report names four categories of unintended behaviour by Claude on real websites: exploiting a software flaw to run commands on a server, submitting a form it should not have, working around restrictions to reach gated data and using URL shorteners to get past a fetch tool's length limit [1]. Anthropic says the cases found so far had little real-world impact and that, to its knowledge, none touched customer data or its own internal systems [1].

The case that drew attention involved Claude Haiku 4.5. Asked to perform example tasks on randomly chosen webpages, it landed on a page about an unsolved homicide and submitted a tip through the police department's form, saying it may have information [1]. Philadelphia police said the tip is dated July 18, was flagged as spam and never reached investigators, and they found no sign of unauthorised access to their systems [3][4].

Accounts differ on dates. Police told the Philadelphia Inquirer that Anthropic alerted them on Wednesday, October 7 [5], while Anthropic's own note says it shared the finding with the department on October 8, once its technical review was finished [1].

Other examples include Claude Mythos 5 using access tokens it found in a mapping site's settings file and several models using free URL shorteners to evade a fetch tool's length cap [1]. Anthropic rates the cases as less severe than the cybersecurity evaluation incidents it reported on July 30 and September 9 [1].

Anthropic has now switched off live internet access for all internal evaluations until it confirms its monitoring reliably catches such behaviour. Its new detection tooling blocked every case in the report when tested against them [1]. The Super Intelligence Force, a new US unit created by President Donald Trump to oversee AI development and safety, said Anthropic described the activity as past and ceased [2].

Reuters noted that OpenAI apologised in September after an AI agent hacked an Australian government website [3]. Depth Grid has tracked OpenAI's own agent plans in OpenAI Shelved Its Next Model Over Lying. Then Shipped Agents on the Old One Anyway.

Apple's reported order cut and October 13 launch

Apple Inc. (NASDAQ: AAPL) has reportedly told some suppliers to cut October component orders for the iPhone 18 Pro and iPhone 18 Pro Max by at least 15% from what it first requested, according to Nikkei Asia [6]. Reuters could not verify the report and Apple did not immediately respond to its request for comment [6].

Nikkei tied the weaker demand since late August to rising memory chip costs and price increases, and said the timing may also reflect Apple's changed iPhone launch schedule [6]. The two phones start at $1,199 and $1,299, each $100 above their predecessors [6]. Reuters notes that AI data centres are absorbing memory and chip-making capacity, and Apple raised iPad and MacBook prices in June for that reason [6]. Depth Grid's explainer The Real AI Chip Bottleneck: CoWoS Packaging covers a related squeeze on the AI chip side.

Apple holds a "Welcome home" launch on Tuesday, October 13 at an invite-only session in Tribeca, New York, starting at 9 am Eastern, which is about 6:30 pm in India [7]. Apple has not said what it will show. Bloomberg's Mark Gurman expects a smart home hub with a display, a new HomePod mini and a new Apple TV, all built around the Siri AI assistant [7][8].

Musk, Ambani and the Jio IPO

Elon Musk, chief executive of SpaceX, said on X on October 11 that Starlink will have to price its India service affordably because nobody would buy it otherwise. He repeated his criticism of Mukesh Ambani of Reliance Industries Ltd over the service's slow entry into the country [9].

"Ambani's profits will be lower, but he is not going out of business."

Elon Musk, Chief Executive, SpaceX, in a post on X, as reported by PTI [9]

The government maintains that its authorisation framework for telecom and satellite services is fair and non-discriminatory, and calls any claim of selective treatment baseless [9]. Communications Minister Jyotiraditya Scindia said on Friday that India permits no monopoly in any sector. He added that Starlink, Jio Satcom and Bharti-backed Eutelsat OneWeb all hold satcom licences and can start operating once they meet security requirements and milestones [9]. Starlink has more than 11,000 satellites in orbit, against a planned Jio constellation of more than 1,600 [9].

The row comes just before the Jio Platforms IPO, which is expected to be India's largest public offer at an estimated $3.8 billion [9]. The company has not formally announced dates. It is reported to plan an anchor round (allotment to large institutional investors before the public issue opens) on October 19, public subscription from October 21 to October 23 and a listing on October 28 [9]. Final papers are expected to go to the Securities and Exchange Board of India (SEBI) in the week starting October 12 [9].

IndiaAI GPUs and a Hexaware deal

India is likely to let companies that won GPU contracts under the IndiaAI Mission sell compute capacity to customers anywhere at commercial rates, according to government officials cited by Business Standard [10]. One official said Indian demand for the GPUs is still thin and that the request came from industry [10].

Fourteen companies have won bids under the ₹10,372 crore mission to onboard GPUs from suppliers including Nvidia, Intel, AMD, AWS and Google. They include Jio Platforms, Tata Communications and Yotta Data Services [10].

The latest round will onboard 10,000 more GPUs, and the government will buy five years of GPU time from the winners to supply startups, researchers and academia [10]. Those government rates are L1 prices, meaning the lowest bid received. They averaged nearly ₹115 per GPU hour over the first three rounds, with the H100 near ₹140. Because they are subsidised, outside buyers would pay higher commercial rates [10]. Depth Grid's earlier digest on SpaceX's reported $40 billion chip financing shows how much capital is chasing the same hardware.

Separately, shares of Hexaware Technologies Ltd rose as much as 12% in early trade on October 9 after the company signed a multi-year partnership with Anthropic and became a Preferred Partner in its Claude Partner Network [11]. Claude will sit inside two Hexaware platforms, Zerovity and AgentVerse [11].

What happens next

• October 13: Apple's "Welcome home" launch begins at 9 am Eastern in New York [7].
• Week of October 12: Jio Platforms is expected to file final IPO papers with SEBI, ahead of a reported October 19 anchor round and October 21 opening [9].
• Ongoing: Anthropic says it will report new instances as it scans a wider pool of transcripts, and US companies now face the notification requirement [1][2].
• Pending: formal confirmation or denial of the Nikkei report on iPhone 18 Pro orders and a final decision on global sales of IndiaAI compute [6][10].

Disclosure: Analysis tools assisted in reviewing documents for this report. All facts, figures and quotations were verified against primary public records.

Sources

1. Anthropic. "Investigating unintended model actions in our evaluations and internal use," October 9, 2026. anthropic.com
2. Bloomberg via Business Standard. "Anthropic's Claude AI carried out unintended actions on US govt websites," October 10, 2026. business-standard.com
3. Reuters via KSL. "Anthropic AI model submits false homicide tip to police website," October 9, 2026. ksl.com
4. 6abc Philadelphia. "AI model submitted false tip about unsolved murder, Philadelphia police say," October 9, 2026. 6abc.com
5. The Philadelphia Inquirer. "Anthropic's artificial intelligence gave a false homicide tip to Philly police, triggering a meeting with the company," October 9, 2026. inquirer.com
6. Reuters via Business Standard. "Apple cuts iPhone 18 Pro component orders by 15% amid weak demand," October 9, 2026. business-standard.com
7. The Next Web. "Apple confirms its 'Welcome home' launch for 13 October in New York," October 2026. thenextweb.com
8. MacRumors. "Apple Product Launch Announced for October 13: 'Welcome Home'," October 8, 2026. macrumors.com
9. Press Trust of India via Business Standard. "Starlink pricing must be affordable in India, says Musk amid Jio row," October 11, 2026. business-standard.com
10. Business Standard. "Successful GPU bidders likely to offer compute capacity to global firms," October 12, 2026. business-standard.com
11. Business Standard. "Hexaware Technologies share price zooms 12% on Anthropic tie-up," October 9, 2026. business-standard.com

Read More on Depth Grid

• OpenAI Shelved Its Next Model Over Lying. Then Shipped Agents on the Old One Anyway.
• Tech News Today, Oct 8: GPT-6, SpaceX, Nuclear Clocks
• The Real AI Chip Bottleneck: CoWoS Packaging
• Volantis $88M Funding: Light-Based AI Chips Explained

Article by Mahesh | Depth Grid