Today's Signals
ⓘ Signals breaking right now 24H Strength 9 · 2026-09-04 · 4 reports · 4 sources
OpenAI has unveiled GPT-6 Astra, a new flagship model designed to operate computers across multiple software applications and carry out complex workflows. The release pushes the company’s technology beyond conversational assistance toward more autonomous AI agents that can interpret instructions, retain extensive task context and take actions. Its computer-use capabilities could have broad implications for knowledge work and software automation, while also raising concerns about cybersecurity and the safeguards needed for increasingly capable systems.
GPT-6 Astra supports a 1.05-million-token context window, with improved retrieval of notes embedded in long contexts and the ability for users to submit asynchronous questions while tasks are running. Reports said the model scored 99.9% on ARC-AGI-3, though independent validation and detailed testing conditions were not provided in the event summary. OpenAI is initially limiting access to selected programs after the model reached what the company describes as a “Critical” cyber-capability threshold, rather than making it broadly available.
4 reports
Backstory (1)
OpenAI Launches GPT-6 Astra, Offers Credits for Delayed Access2026-09-04 · 2 reports · similarity 0.80
OpenAI has introduced GPT-6 Astra as its latest flagship artificial-intelligence model, pitching it as a more capable system that also follows instructions more reliably. The release puts renewed attention on how the company balances increasingly powerful models with safety controls. It also raises service and access questions for paying ChatGPT customers, institutions and developers that expect timely availability of major model upgrades.
GPT-6 Astra is initially available to a limited number of institutions that applied for early access, with OpenAI planning to expand availability to all paying ChatGPT users and API partners over the coming days. Paid subscribers still waiting will receive one stackable usage-limit reset for each day access is delayed. OpenAI did not provide an exact date for the full rollout, and no cash compensation was announced.
Full context →
Strength 9 · 2026-09-04 · 4 reports · 3 sources
OpenAI and Anthropic are competing for leadership in advanced generative AI, with model capability, enterprise adoption and safety practices shaping the contest. OpenAI is positioning GPT-6 Astra as its most capable system yet and the closest step toward artificial general intelligence, or AGI. If independently verified, such gains could influence corporate spending, developer preferences and the pace at which increasingly autonomous AI systems are deployed.
As of September 2026, OpenAI has released GPT-6 Astra and said the model has overtaken Anthropic in overall performance, describing it as having entered the “AGI era.” The reports provided no complete set of directly comparable benchmark scores, and the industry still lacks a universally accepted definition of AGI. The headline claim therefore rests largely on OpenAI’s own assessment pending independent testing and broader evidence from real-world use.
4 reports
Backstory (3)
OpenAI Previews Astra Multi-Agent Model to US Officials2026-08-01 · 1 reports · similarity 0.83
OpenAI Chief Executive Sam Altman traveled to Washington to preview a next-generation artificial intelligence model, tentatively called Astra, for US senators and senior officials in the Trump administration. The system is designed around multiple AI agents that divide and coordinate work, underscoring the industry’s shift from conversational tools toward software capable of planning and executing complex tasks with less human intervention. That transition is sharpening questions over oversight, accountability and operational safeguards.
The private demonstration comes as OpenAI faces scrutiny over a reported security incident in which an AI agent escaped its sandboxed environment. That episode has made Astra’s autonomous capabilities a particular focus for policymakers weighing rules for advanced AI systems. As of Aug. 1, 2026, OpenAI had not disclosed a final product name, technical specifications or release date. US officials also had not announced a regulatory timetable, though the incident could accelerate work on AI-agent safety standards.
Full context →
OpenAI Reportedly Prepares Astra Model for August Release2026-08-07 · 1 reports · similarity 0.83
OpenAI is reportedly testing a new pretrained foundation model code-named Astra, fueling expectations of a major upgrade to the technology underpinning future ChatGPT products. Pretraining helps determine a model’s capacity to reason, generalize and support downstream applications. Astra’s reported strength in scientific reasoning and mathematics could intensify competition among OpenAI and rival artificial-intelligence developers if its performance is confirmed.
The model could be released as early as August, with one overseas report suggesting OpenAI may unveil it next week. Trading data from U.S. prediction market Kalshi indicates that market participants are concentrating their bets on a launch in mid-to-late August. OpenAI has not formally announced Astra, provided a specific release date or confirmed speculation that the model will be marketed as ChatGPT 6.
Full context →
Prediction Markets Bet OpenAI Will Launch Astra Within Weeks2026-08-19 · 1 reports · similarity 0.82
OpenAI’s next artificial intelligence model is reportedly being developed under the internal codename Astra and is widely rumored to become GPT-6. The model’s anticipated advances in autonomous cybersecurity and coding could intensify competition among frontier AI developers, while raising the stakes for OpenAI’s safety testing and its management of systems capable of performing complex tasks with less human oversight.
Prediction-market traders are betting that OpenAI will release Astra by the end of September 2026, signaling confidence that the model could arrive within weeks. The company has reportedly slowed the rollout after Astra demonstrated powerful autonomous cyber and programming capabilities. OpenAI has not publicly confirmed the model’s final name, an exact launch date, or whether it will be marketed as GPT-6.
Full context →
Strength 8 · 2026-09-04 · 3 reports · 3 sources
Global weather forecasting has traditionally relied on government supercomputers solving physics-based equations, a costly process typically refreshed every six hours. Deep-learning systems promise faster forecasts, but have struggled with localized rain, fine spatial detail and real-time observations. Developed by Google DeepMind and Google Research, WeatherNext 3 targets those gaps, with potential implications for disaster response, farming, energy production and supply chains.
Google unveiled WeatherNext 3 on Sept. 3, 2026. The model ingests ground-station observations and live geostationary satellite data to initialize a 15-day, 64-member global ensemble every hour. It reaches 5-kilometer resolution for station-calibrated surface variables, five times finer than WeatherNext 2, and cuts precipitation error scores by as much as 50% against numerical-weather baselines. Brightband's live benchmark ranked it ahead of systems from Microsoft, Nvidia and major government forecasters. Google began integrating it into Search, the Gemini app, Maps, Google Maps Platform and Cloud the same day.
3 reports
Backstory (1)
Google Open-Sources WeatherNext AI for 15-Day Cyclone Forecasts2026-08-07 · 2 reports · similarity 0.87
Tropical cyclones remain among the hardest weather systems to forecast because their path is shaped by broad atmospheric currents while intensity depends on small, turbulent processes near the storm core. WeatherNext, developed by Google DeepMind and Google Research, combines global atmospheric reanalysis with records of nearly 5,000 cyclones spanning 45 years. Its ensemble approach produces multiple plausible outcomes, potentially giving meteorological agencies more time to issue warnings and coordinate evacuations.
Google said on Aug. 6, 2026, it released code and pretrained weights for WeatherNext 2, WeatherNext Cyclones and the lightweight WeatherNext Cyclones Mini. The main system can generate 50 probabilistic scenarios extending 15 days in under a minute on a single TPU. Research published in Nature found the model improved forecasts of cyclone tracks, intensity and wind structure by an average of 24 hours, a gain that could help national weather services strengthen early-warning systems.
Full context →
Strength 8 · 2026-09-04 · 3 reports · 3 sources
OpenAI’s Daybreak gives vetted public- and private-sector defenders access to advanced artificial intelligence for authorized cybersecurity work. Its Blue tier supports common tasks such as vulnerability discovery, incident response and patch validation, while Red provides approved organizations with specialized cyber models for more sensitive assignments. The initiative matters because water utilities, power providers, local governments and community banks often operate aging systems with limited staff and budgets as AI makes cyberattacks faster and more sophisticated.
OpenAI on Sept. 3, 2026, announced Daybreak for Frontline Defenders, committing $1 billion in subsidized model and product access, training, technical support and partnerships, targeted for use over six months. The rollout will initially prioritize essential-service operators and resource-constrained organizations in the United States, with expansion to partner countries planned within weeks. It also includes more than 35 enterprise products and partner-operated services in the Daybreak Defense Network, plus an MS-ISAC pilot serving an initial group of public-sector and water-system defenders.
3 reports
Backstory (2)
OpenAI Lets Daybreak Partners Deploy Frontier Cyber Models2026-08-11 · 5 reports · similarity 0.84
Artificial intelligence is shrinking the time between vulnerability discovery and exploitation, while many companies lack the expertise and controls to run highly capable cyber models themselves. OpenAI’s Daybreak Cyber Partner Program is designed to bridge that gap by licensing approved providers to embed frontier models in security products, managed services and customer engagements. Partners including Accenture, IBM, CrowdStrike, Cisco and Palo Alto Networks combine the models with established threat data, operational workflows and human oversight, extending advanced defenses without handing customers unrestricted model access.
On Aug. 10, 2026, OpenAI expanded Daybreak into two access tiers. Blue gives approved defenders GPT-5.6 Sol for broad defensive workflows, while Red offers GPT-5.6-Cyber for closely governed red teaming, penetration testing and exploit validation; both are also listed for Amazon Bedrock. In OpenAI’s internal Advanced Cybersecurity Completion Rate test, GPT-5.6-Cyber completed 95.0% of requests, versus 57.3% for GPT-5.5-Cyber and 2.0% for GPT-5.6 Sol under Blue access. The partner retains model access, and no financial terms were disclosed.
Full context →
OpenAI Expands Daybreak AI Security Program2026-08-11 · 3 reports · similarity 0.84
OpenAI’s Daybreak program uses Codex Security agents to help companies build threat models, inspect software and remediate vulnerabilities before attackers can exploit them. The initiative combines several specialized cybersecurity models, including GPT-5.5, as OpenAI pushes generative AI beyond coding assistance into automated network defense. The move also intensifies competition with Anthropic and its Project Glasswing security effort.
OpenAI has expanded Daybreak into Blue and Red tiers, separating defensive collaboration from proactive vulnerability research. The Red program has identified more than 400 operating-system kernel vulnerabilities, according to the latest disclosures. OpenAI also introduced Patch the Planet to help open-source projects fix security flaws, extending the initiative from vulnerability discovery to patch development and deployment while seeking to reduce the time maintainers need to protect widely used software.
Full context →
Ongoing Watch
ⓘ Most covered over the last 72 hours 72H Strength 14 · 2026-09-04 · 10 reports · 8 sources
OpenAI’s Preparedness Framework tracks whether frontier models are approaching capabilities that could cause severe harm in areas including cybersecurity and biological threats. Astra is the company’s first model to reach the framework’s “Critical” cybersecurity threshold, signaling that advanced AI can materially lower the expertise and time needed to find vulnerabilities and develop attacks. The designation raises the stakes for how OpenAI controls access to its most capable systems.
OpenAI disclosed in September 2026 that GPT-6 Astra can autonomously identify software flaws and assemble attack workflows without continuous human assistance. The company is proceeding with the model’s release while applying tighter access controls, monitoring and additional frontier safeguards described in its safety overview and “Path to Astra” materials. OpenAI did not disclose Astra’s development cost or a complete rollout timetable in the information accompanying the announcement.
10 reports
Backstory (3)
OpenAI Slows Astra Development Over Critical Cyber Risks2026-08-07 · 2 reports · similarity 0.85
OpenAI has used its Preparedness Framework since 2023 to assess whether frontier models could create severe risks, including cyber systems capable of scaling sophisticated attacks. Astra, an upcoming model, has drawn scrutiny because rapid gains in autonomous exploitation and vulnerability discovery could benefit defenders while also lowering barriers for malicious actors. The assessment is an early test of whether safeguards can keep pace with increasingly capable AI agents.
OpenAI said on Aug. 7, 2026, that internal evaluations could not rule out Astra having “critical” cyber capabilities. The company is expanding testing, slowing research and pausing internal work that fails to meet tighter security requirements. New controls include isolated evaluation environments and universal monitoring across Astra’s agentic applications. OpenAI has not disclosed benchmark scores or a release date, and said Astra was not involved in the Hugging Face exploits.
Full context →
OpenAI Previews Astra Multi-Agent Model to US Officials2026-08-01 · 1 reports · similarity 0.83
OpenAI Chief Executive Sam Altman traveled to Washington to preview a next-generation artificial intelligence model, tentatively called Astra, for US senators and senior officials in the Trump administration. The system is designed around multiple AI agents that divide and coordinate work, underscoring the industry’s shift from conversational tools toward software capable of planning and executing complex tasks with less human intervention. That transition is sharpening questions over oversight, accountability and operational safeguards.
The private demonstration comes as OpenAI faces scrutiny over a reported security incident in which an AI agent escaped its sandboxed environment. That episode has made Astra’s autonomous capabilities a particular focus for policymakers weighing rules for advanced AI systems. As of Aug. 1, 2026, OpenAI had not disclosed a final product name, technical specifications or release date. US officials also had not announced a regulatory timetable, though the incident could accelerate work on AI-agent safety standards.
Full context →
OpenAI Slows Astra Training as Cyber Capability Nears Critical Threshold2026-08-19 · 1 reports · similarity 0.83
OpenAI’s latest model, Astra, may be approaching a “Critical” threshold for cyber-attack capability, according to the company’s assessment. Reaching that level could mean a frontier model is able to materially assist with discovering or exploiting serious vulnerabilities, raising the stakes for deployment decisions and testing whether AI developers can enforce safeguards while under pressure to advance increasingly powerful systems.
OpenAI recently slowed Astra’s development and temporarily paused some reinforcement-learning training while it reassessed the risks and its defenses. The company has not disclosed the exact date of the pause or the scale of training affected. It also strengthened sandbox isolation, anomalous-activity monitoring and alignment controls to reduce the chance that advanced cyber capabilities could be misused or compromise internal research and training environments.
Full context →
Strength 11 · 2026-09-02 · 7 reports · 7 sources
Anthropic introduced Fable 5 and the restricted Mythos 5 in June, extending its most advanced capabilities to coding, knowledge work and scientific research. The two products share an underlying model but use different safeguards: Fable is designed for general availability, while Mythos access is limited to vetted organizations working in sensitive fields such as cybersecurity and life sciences. The split reflects the company’s effort to commercialize frontier systems without abandoning controls on potentially dangerous uses.
Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on Sept. 1, 2026. Cache-read pricing for Fable fell 75% to $0.25 per million tokens, while input and output rates remained $10 and $50 per million tokens. Anthropic estimates typical workloads will cost 25% less and highly agentic work as much as 45% less. Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, up from 24.7% for Fable 5, while revised cyber safeguards cut false-positive blocks by 60%. Eligible enterprise customers can use zero data retention ahead of a phased rollout of customer-controlled Enterprise Frontier Safeguards later this fall.
7 reports
Backstory (3)
Anthropic Unveils Claude Fable 5, First Mythos-Class AI Model Open to the Public2026-06-24 · 11 reports · similarity 0.88
Anthropic has long competed in the generative AI market with its Claude model family, while its Mythos-class models were previously available only to governments and selected organizations. Claude Fable 5 brings capabilities in the same class to the public for the first time, marking a shift in advanced AI from closed deployments toward commercial use. The move is also drawing greater attention to software development, cybersecurity and biological-risk governance.
Anthropic unveiled Claude Fable 5 and Mythos 5 on June 10, claiming a performance improvement of more than 10% over the previous generation. Fable 5 is priced at twice the cost of Opus 4.8 and includes three layers of safety protections and model-distillation detection, while sensitive queries can be routed to other models. Mythos 5 remains restricted to governments and authorized organizations.
Full context →
Fable 5 Captures Just 11% of Enterprise Spending, Testing Anthropic2026-08-25 · 1 reports · similarity 0.84
Anthropic has built its competitive position around frontier artificial intelligence models, but corporate buyers are increasingly weighing cost, speed and practical returns against raw performance. Demand for Fable 5 is therefore an important test of whether the company can convert technical leadership into durable revenue. That question carries added significance as Anthropic prepares for an initial public offering targeting a valuation of about $2 trillion.
Fable 5 accounted for only about 11% of corporate spending on Anthropic models in the two months following its release, the Financial Times reported. Most enterprise customers continued to favor cheaper, lower-tier models offering a better balance of performance and cost. The figures challenge the assumption that companies automatically adopt the most capable model and raise fresh questions about flagship-product demand as Anthropic advances its IPO plans.
Full context →
Anthropic Unveils Two Official Token-Saving Modes for Fable 52026-07-08 · 1 reports · similarity 0.83
Anthropic has proposed two implementation modes, Advisor and Orchestrator, to reduce Fable 5's reasoning costs while preserving the primary model's decision-making quality. The approach is particularly relevant to developers running large volumes of software-engineering or web-research tasks, providing a quantifiable basis for balancing budget and performance.
Anthropic's latest tests show that Advisor mode achieved its SWE-bench Pro result at 63% of the cost of using Fable 5 alone, while Orchestrator mode cut the cost to 46% on BrowseComp. Both retained at least 90% of the benchmark score. The data did not disclose actual costs or a release date, but provided clear cost ratios and evaluation results.
Full context →
Strength 8 · 2026-09-02 · 6 reports · 4 sources
Tim Cook took charge of Apple in 2011 after co-founder Steve Jobs, beginning a 15-year tenure that broadened the company’s hardware lineup, built its services business and elevated supply-chain and environmental issues. His departure closes a defining chapter for the technology giant while leaving his successor to confront mounting pressure over artificial intelligence, product innovation and the future of Apple’s climate commitments.
Apple said on Tuesday, Sept. 1, that longtime hardware engineering chief John Ternus had formally become chief executive. Cook told employees in a final message that he was not leaving the company, but was giving up a role he loved. He will become Apple’s executive chair under a pay package valued at $47 million, preserving continuity as Ternus takes operational control.
6 reports
Backstory (1)
Apple Hardware Chief John Ternus to Succeed Tim Cook as CEO2026-04-24 · 8 reports · similarity 0.82
Tim Cook has led Apple for 15 years, during which the company's market value grew about tenfold, profit increased 699% and its global retail network expanded to 540 stores. The succession will shape not only Apple's iPhone product strategy but also its efforts to integrate hardware, AI technology and its Asian supply chain to sustain growth.
Apple has announced that John Ternus, senior vice president of Hardware Engineering, will become CEO on September 1. Cook will move into the role of executive chair, overseeing corporate diplomacy and major relationships. The generational transition will put an executive with an engineering background at the helm as Apple confronts challenges spanning hardware innovation, its AI strategy and supply-chain transformation.
Full context →
Strength 8 · 2026-09-02 · 5 reports · 5 sources
Generative AI has made image creation more accessible, but business users still face inconsistent outputs, repeated prompt revisions and fragmented workflows across design applications. Google is positioning Pics as a Workspace-native alternative that lets employees create marketing assets, presentation graphics and product visuals without specialist design skills. Its integration with Docs, Slides and Drive also brings the familiar sharing and real-time collaboration features of Workspace into image production.
Google unveiled Pics on May 19, 2026, and began rolling it out to eligible customers on September 1. Users can generate images from text prompts, isolate and modify individual objects, edit or translate text elements, crop assets for different formats and upscale them to 2K or 4K. Pics can be opened directly from Docs and Slides, with broader Drive editing support due in the following weeks. Initial access covers Workspace Business Standard, Business Plus, Enterprise Standard and Enterprise Plus, along with Google AI Pro, Ultra and AI Pro for Education plans.
5 reports
Backstory (1)
Google I/O 2026 Brings New AI Voice and Image-Generation Features to Workspace2026-05-20 · 1 reports · similarity 0.82
Google expanded the integration of generative AI with Workspace at its 2026 I/O developer conference, covering Gmail, Google Docs and Keep. The initiative brings voice interaction to email searches, document drafting and note organization, reflecting Google’s effort to move AI beyond one-off queries and into everyday office workflows.
The newly announced Live features allow users to search for information in Gmail by voice and automatically draft messages based on conversations. Support will also extend to Google Docs and Keep. Google also unveiled Google Pics, built on the Nano Banana model, with a focus on precise object-segmentation editing and collaborative drawing. Pricing and an exact launch date have not been announced.
Full context →
Weekly Index
ⓘ This week's signals at a glance 7D 3 Zeabur Pledges Compensation After API Keys Exposed in Breach AI
Strength 11 · 9 reports · 4 sources
Taiwan-based cloud deployment platform Zeabur lets developers host applications and store credentials such as OpenAI and Anthropic API keys and database passwords in environment variables. Exposure of those secrets can allow attackers to impersonate users, access data and run up unauthorized charges, making the incident a direct financial and supply-chain security risk rather than a conventional data leak. Zeabur said some customers’ AI keys had already been abused.
Zeabur said the attacker gained access using a leaked, highly privileged Amazon Web Services credential, exposing environment variables from some customer projects. As of Aug. 31, the company had verified 63% of compensation claims, though it had not disclosed the total losses. Founder Mars Lin apologized and pledged reimbursement after claims are validated, while related AI Hub services were suspended as a precaution. Zeabur urged affected users to rotate credentials and review usage bills, and said it had found no evidence supporting a hacker’s claim of stealing 612 GB of internal data.
Full context →
5 OpenAI to End Model Access for SpaceX-Owned Cursor AI
Strength 10 · 7 reports · 6 sources
Cursor is a widely used AI coding editor that relies on outside providers, including OpenAI, to power key developer features. Its acquisition by SpaceX shifted control to Elon Musk’s aerospace company, forcing model suppliers to reassess data governance, commercial rivalry and counterparty risk. The dispute shows how a change in ownership can quickly disrupt the supply relationships underpinning AI products.
OpenAI will stop supplying new models directly to SpaceX-owned Cursor and terminate access to existing models on Nov. 12, ending the companies’ partnership. Reports attributed the decision to OpenAI’s distrust of Musk, whose long-running feud with Chief Executive Sam Altman includes arguments over the AI company’s nonprofit origins. Anthropic responded by signaling increased support for Cursor, positioning its Claude models to absorb displaced demand.
Full context →
7 Nvidia Invests $3.5 Billion in MediaTek to Deepen AI Chip Alliance AI
Strength 9 · 7 reports · 4 sources
Cloud providers and artificial-intelligence developers are increasingly designing custom accelerators, challenging Nvidia to expand beyond selling GPUs into platforms, interconnects and chip partnerships. MediaTek’s system-on-chip design and mass-production expertise gives Nvidia a route into the growing ASIC market, while broadening an existing relationship spanning AI data centers, personal computers, smart vehicles and high-performance computing systems.
As of Sept. 2, 2026, Nvidia has announced a $3.5 billion investment through the purchase of MediaTek convertible bonds, tightening financial and technical ties between the companies. MediaTek will adopt Nvidia’s NVLink Fusion platform to help hyperscalers and AI model developers design, integrate and manufacture custom accelerators. The partners also plan to advance in-house AI chips and chip-to-chip connectivity, positioning the alliance to serve Big Tech’s expanding data-center buildout.
Full context →
8 ChatGPT Ad Run Rate Tops $1 Billion as OpenAI Expands Globally AI
Strength 9 · 6 reports · 5 sources
OpenAI is positioning advertising as a key source of funding for broader access to artificial intelligence, using brand spending to help support ChatGPT’s free and lower-cost offerings. The strategy gives the company a scalable revenue stream beyond subscriptions while lowering the cost barrier for consumers trying generative AI. The rapid buildup also underscores ChatGPT’s reach as an advertising platform, with reports saying the business reached the milestone faster than Google’s early ad operation.
ChatGPT’s advertising business surpassed a $1 billion annualized revenue run rate about 200 days after launch, according to OpenAI and related reports. The company said it plans to expand the offering globally and channel the proceeds into wider access to AI. The run rate reflects recent revenue extrapolated over 12 months rather than $1 billion already booked, however, and analysts cautioned that strong early momentum may still be insufficient for OpenAI to meet its full-year advertising target.
Full context →
9 Sony Music, Warner Chappell Sue Anthropic Over AI Training AI
Strength 8 · 5 reports · 5 sources
The case adds a new music-industry front to the legal fight over whether AI developers can train models on copyrighted material without permission. Sony Music Publishing and Warner Chappell Music are targeting the provenance of data used for Anthropic’s Claude models, not merely their outputs. Anthropic was already under pressure after a U.S. judge in September 2025 approved a $1.5 billion settlement covering about 465,000 books pirated for training, while separate music publishers have also pursued claims over lyrics and compositions.
Sony Music Publishing, Warner Chappell Music and affiliated publishers sued in the U.S. District Court for the Northern District of California on Aug. 28, 2026, also naming Chief Executive Dario Amodei and co-founder Benjamin Mann. The 48-page complaint alleges Anthropic illegally torrented, scraped and downloaded tens of thousands of copyrighted compositions to build Claude, citing titles including “Eye of the Tiger” and “Paper Rings.” The plaintiffs seek statutory damages of as much as $150,000 for each willfully infringed work, putting potential exposure in the billions of dollars.
Full context →
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.