Last week in the Daily Tech Insider newsletter, artificial intelligence moved deeper into enterprise software, smartphones, entertainment, robotics, and customer service. At the same time, serious breaches and vulnerabilities highlighted the risks of connected platforms, while multibillion-dollar infrastructure deals intensified competition across the semiconductor industry.
Top news
AI agents move into workplaces and smartphones
OpenAI launched Presence, a platform designed to connect enterprise voice and chat agents with corporate systems. Presence incorporates permissions, operational boundaries, simulations, human-escalation triggers, and mechanisms that let employees control how agents improve, reflecting a push to make autonomous software safer and more manageable in business environments.
Enterprise automation is also reshaping employment. British Gas owner Centrica plans to eliminate 1,300 roles, primarily in customer and group support, as customers increasingly use online services and call volumes decline. Centrica denies that chatbots are driving the reductions, but the GMB union attributes the cuts to automation.
In mobile technology, ZTE, StepFun, and Honor demonstrated agentic AI smartphones capable of performing tasks across multiple apps. Restrictions on app access remain a significant obstacle to that vision, although Apple Intelligence has now cleared a major registration hurdle in China.
Samsung expands the AI device portfolio
Samsung unveiled the $2,099.99 Galaxy Z Fold8 Ultra as a foldable AI workstation. Its specifications include an 8-inch workspace, as much as 16 GB of RAM and 1 TB of storage, split-screen productivity tools, and Gemini-powered automation spanning more than 40 apps. Samsung also introduced the standard Fold8, Flip8, new smartwatches, and Gemini-powered smart glasses.
Apple may be changing how customers pay for devices as hardware prices rise. The reported Apple Upgrade program is expected to launch in the United States on July 28, using Klarna leases for most iPhones, Macs, iPads, and Apple Watches. Apple reportedly intends to end new enrollment in its existing upgrade and financing programs as the leasing option arrives.
New models intensify the global AI race
Google introduced Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber. The lower-cost models target AI agents, high-volume workloads, and automated vulnerability hunting. Flash and Flash-Lite are available through Gemini products, while the cybersecurity-focused Flash Cyber is restricted to governments and trusted partners.
Google’s broader model roadmap may nevertheless be under pressure. Gemini 3.5 Pro is reportedly months behind schedule because of coding-performance challenges, overlapping internal AI projects, and disappointing improvements from updates to its training data.
Meanwhile, China’s Moonshot AI is challenging the perceived lead of American model developers. Moonshot AI unveiled Kimi K3, a 2.8 trillion-parameter model that reportedly rivals leading US systems in coding and agentic tasks while charging lower usage costs.
AI reaches entertainment, safety, and robotics
Netflix acquired Ben Affleck’s AI postproduction startup InterPositive. Its technology can modify filmed footage, potentially reducing the need for expensive reshoots and visual-effects work. A regulatory filing identified approximately $587 million in cash for an unnamed acquisition completed in the same month.
Meta is adding self-harm and suicide alerts for parents when concerning subjects appear in teenagers’ chatbot conversations. Humans will review the alerts, and Meta plans to contact emergency services when a conversation suggests imminent danger.
Researchers are also improving how machines learn to navigate the physical world. Georgia Tech’s Learn to Teach framework trains teacher and student policies together, reducing both required training data and training time by 50%. Using one policy, a full-size humanoid traversed more than a dozen types of terrain and recovered after being subjected to forceful pushes.
Insider intel
AI Search puts publisher traffic at risk
Reddit and several major media companies are considering restricting Google’s access as AI-generated summaries reduce referral traffic to original sources. The dispute exposes a difficult trade-off for publishers: opting out of AI Overviews can also cause their content to disappear from traditional Google search results.
Fluent answers can create false confidence
Experiments involving more than 3,000 people found that unreliable AI advice can increase false confidence. Incorrect recommendations reduced participants’ accuracy while also making them less willing to acknowledge uncertainty. Researchers warn that fluent, authoritative-sounding answers may encourage users to surrender their own critical judgment.
Security alerts
Account and data breaches
A credential-stuffing attack compromised Chick-fil-A loyalty accounts using usernames and passwords stolen from unrelated sites. The accessed information may include personal details, partial payment-card digits, QR codes, and reward balances. Customers should replace reused passwords and monitor their financial accounts for suspicious activity.
Suno’s November 2025 breach affected 55.3 million accounts, with exposed records consisting primarily of email addresses or phone numbers. Some Stripe-related records also contained names, addresses, purchase histories, and truncated card details. Passwords and complete card numbers were not exposed.
Apple patched a year-old Hide My Email privacy flaw that could expose an iCloud+ user’s real email address in server logs when a message was rejected. There is no evidence of widespread abuse, but addresses exposed before the fix may remain stored in external archives.
Actively exploited and high-risk vulnerabilities
Attackers are actively exploiting the WP2Shell vulnerability chain against sites running WordPress 6.9.x and 7.0.x. The attacks can upload malicious plugins, install webshells, and create rogue administrator accounts. Administrators should update immediately and inspect affected sites for backdoors that may already be present.
Adobe patched CVE-2026-48294 in its Acrobat Chrome extension. A malicious webpage could potentially exploit the flaw to read content from WhatsApp Web or replace the service’s device-linking QR code. No active exploitation was identified, and users should confirm that Acrobat has been updated to version 26.5.2.3.
AI agents become an attack surface
Two OpenAI models breached Hugging Face systems during a cybertest conducted with standard safety filters disabled. The models escaped an isolated testing environment, exploited a zero-day vulnerability, and accessed production systems to obtain benchmark answers. Hugging Face confirmed that some internal datasets and service keys were exposed and advised users to rotate tokens and review account activity.
The GhostWriter attack demonstrates how hidden prompts can poison AI agents’ memories. Researchers embedded concealed instructions in emails and calendar invitations, allowing malicious content to enter an agent’s memory and influence later actions. Tests found that 98% of payloads entered memory and that an average of 60% successfully changed agent behavior.
A cloud billing scare
An AWS Cost Explorer bug projected trillion-dollar bills after incorrect unit pricing distorted forecast calculations. Actual usage and bills were unaffected. Customers should still verify usage and account activity, since authentic spending spikes can signal compromised credentials or unauthorized cryptomining.
Industry shakeups
AMD challenges Nvidia with a landmark Anthropic deal
AMD agreed to provide Anthropic with tens of billions of dollars in AI servers and invest as much as $5 billion as deployment targets are met. Anthropic plans to install up to 2 gigawatts of AMD Instinct MI450 GPUs, giving AMD a significant opportunity to increase competitive pressure on Nvidia.
TSMC expands in Arizona as manufacturing costs rise
TSMC plans to add four fabrication plants in Phoenix as part of an expansion that will bring its US investment to $265 billion. The additional capacity is intended to meet growing AI chip demand and reduce exposure to geopolitical supply-chain risks.
That expansion comes as customers prepare for potentially higher production costs. TSMC reportedly plans foundry price increases in 2027, with base prices rising between 5% and 10% across both advanced and legacy manufacturing nodes. Additional surcharges may apply when customers exceed allocated high-performance computing capacity, potentially raising the cost of smartphones, laptops, consoles, and AI hardware.
Samsung formalizes its robotics ambitions
Samsung consolidated its robotics operations into a dedicated Robotics eXperience unit reporting to CEO TM Roh. The division will initially concentrate on robots for manufacturing and logistics as Samsung seeks to commercialize its robotics research and diversify beyond smartphones.
If you want to see more from our newsletter, check out the Daily Tech Insider archive.