Better virtuous than safe


According to the safety principle, if one knows that p, then one’s belief in p could not easily have been false. In this paper, I pose a dilemma for safety theorists by asking the following question: In evaluating whether or not a belief is safe, must we only examine the error-possibilities of the same belief as formed in the actual world? If ‘yes’, safety meets a familiar objection regarding necessary truths and the objection also extends to contingent propositions. If ‘no’, however, there is no necessity of safety for knowledge. It is argued that this dilemma poses a threat for a number of safety principles from the recent literature. In the end, I draw implications of my arguments for the debate between robust and anti-luck virtue epistemologies. The result is that anti-luck virtue epistemology suffers from the difficulty with the safety principle whereas the robust variant remains intact.

This is a preview of subscription content, access via your institution.


  1. 1.

    For example, Sosa (1999) has famously introduced safety as a powerful (neo-Moorean) anti-skeptical condition, and according to him, safety is superior to the sensitivity principle in a number of aspects. Pritchard (2005, 2012) argues that safety captures the intuition that knowledge is incompatible with luck (of a certain kind), and thus he uses safety to deal with the Gettier problem. Other safety theorists include Beddor and Pavese (forthcoming), Broncano-Berrocal (2018), Carter (2010), Greco (2016), Grundmann (forthcoming), Hirvelä (2019), Manley (2007), Peet and Pitcovski (2018), Sainsbury (1997), Whiting (forthcoming) and Williamson (2000).

  2. 2.

    Objections to safety include Baumann (2008), Bernecker (forthcoming), Bogardus (2014), Comesaña (2005), Freitag (2014), Hiller and Neta (2007), Kelp (2009), McEvoy (2014), Neta and Rohrbaugh (2004). For responses to some of these objections, see Beddor and Pavese (forthcoming), Bogardus (2014, pp. 297–299), Broncano-Berrocal (2014), Grundmann (forthcoming), Pritchard (2015) and Williamson (2009a).

  3. 3.

    Thus, for the purposes of this paper, beliefs are individuated at least partially by means of the truth conditions of the propositions that one believes. Same belief entails same truth condition (although same truth conditions do not guarantee same beliefs). For example, the proposition that “Molly is in the room” and the proposition that “A person is in the room” clearly share different truth conditions. For the latter to be true, anyone could be in the room, not just Molly. On this basis, we could say that beliefs in these two propositions are not the same belief.

    Incidentally, there seems to be a kind of “generality problem” with respect to determining “same belief”. A token belief can be typed in different ways, either in a coarse-grained way or a fine-grained way. Thus, when two token beliefs are typed in a very fine-grained way, they may belong to different types of beliefs, but when they are typed in a very coarse-grained way, they may belong to the same type. Therefore, the question remains how token beliefs should be typed in order to determine whether they are of the same type or not. Fortunately, we need not dwell on such a complication. But the problem here may be one that a safety theorist wants to deal with (Cf. Bishop 2010). Thanks to an anonymous referee.

  4. 4.

    See e.g. Manley (2007), Sainsbury (1997), Weatherson (2004) for exceptions.

  5. 5.

    Throughout this paper I am taking ‘method’ broadly as ‘way of belief-formation’. Asking the question of ‘what is your method of believing that p?’ is essentially asking ‘what is your way of believing that p?’ or ‘how do you come to believe that p?’ Most safety theorists agree that safety needs to be restricted to a method in this sense. But depending on one’s overall epistemological leanings, a method may be characterized in different terms (be they internalist, externalist, or virtue-theoretic, etc.). See Nozick (1981, p. 179) for an argument to the effect that method should be held fixed in a modal account of knowledge.

  6. 6.

    Baumann (2008) and Kvanvig (2008) have explicitly pointed out that safety theorists’ methodology of possible world semantics is problematic. Although I am sympathetic with their criticisms, I will not pursue this line of objection in this paper. See also Bogardus (2014) for relevant discussions.

  7. 7.

    See Sainsbury (1997, p. 909) for a similar example.

  8. 8.

    Cf. Hirvelä (2019, p. 1170) who also cites Greco’s example to demonstrate that the problems with SAFETYSAME go beyond necessary propositions.

  9. 9.

    Note that both Mathema and Frog indicate that SAFETYSAME, plus true belief, are insufficient for knowledge. I will return to this point regarding insufficiency in Sect. 6. (See also Vogel (2007), Miracchi (2015), Pritchard (2012) for arguments against the sufficiency of safety for knowledge.)

  10. 10.

    Cf. Manley (2007), Pritchard (2012), Sainsbury (1997) and Weatherson (2004).

  11. 11.

    Exactly how Jack’s method of belief-formation should be individuated is a matter of controversy. Be that as it may, SAFETYDIFF is more plausible than SAFETYSAME, vis-a-vis Frog (Cf. Hirvelä 2019, p. 1172). Notice that SAFETYSAME is unable to explain why Jack fails to know, however one individuates the method.

  12. 12.

    Plausibly, this false belief could also be formed in the actual world, perhaps right after she forms her @belief. If so, the formation of this false belief in the actual world would also be relevant for determining whether or not @belief is safe. For (1) actual world is the closest to itself and (2) both the false belief and @belief are presumably formed via the same method. (I shall further examine whether (2) holds in the next section.).

  13. 13.

    That said, it is important to note here that the spirit of this objection can be equally carried by a safety theorist who does not favor an internalist way of method-individuation. For example, if one favors the externalist method (Williamson 2009a), one could argue that, were Cynthia to believe P2, this belief would be caused by Tim-relevant features, whereas her belief in P1 is not caused by such features. My responses below apply equally well to externalist ways of method-individuation.

  14. 14.

    For example, suppose we describe the method of belief in P1 as “perceiving the two people playing chess while facially recognizing (only) Jay” and the method of belief in P2 as “perceiving the two people playing chess while facially recognizing Jay and Tim”. Now, even if this way of describing the method is granted, it does not apply to the following case which is constructed according to the same recipe of Cynthia’s Perception (not to say that the way of method-individuation here has no implication in the barn-façade case above). Suppose I have no idea that a whale is not a fish. One day, I see a big whale while sailing and I form a belief that it is a big animal. But since I do not know a whale is not a fish, I could easily have formed a false belief that it is a big fish. In spite of such error-possibilities, intuitively I still know that the object is a big animal. Since forming beliefs that the object is a fish/animal has nothing to do with facial recognition, the suggested way of method-individuation does not apply.

  15. 15.

    See Broncano-Berrocal (2014), Greco (2016), Grundmann (forthcoming), Williamson (2009a) for proposals of method-individuation on behalf of the safety theorists. See Bogardus and Marxen (2014), Hirvelä (2018), Zhao (forthcoming) for criticisms.

  16. 16.

    Cf. Williamson (2009b, p. 325), Lasonen-Aarnio (2010, p. 3), Sosa (2015, p. 123). See also Bernecker (forthcoming) for an interesting (testimonial) counterexample to SAFETYSIMILAR. His example, in some way, can be considered as a testimonial counterpart of a case like Cynthia’s Perception.

  17. 17.

    Bernecker (2010, pp. 222–229) argues that propositional similarity, in the context of propositional memory, should be understood in terms of relevant entailment (See also Bernecker forthcoming, p. 10).

  18. 18.

    An anonymous referee points out that such authors as Williamson (2009a) have adopted safety within a knowledge-first framework, where safety serves as a circularly necessary condition for knowledge. That is, for these theorists, pre-theoretic intuitions about knowledge can be used to make judgments about safety, which implies that they could use their intuitions about knowledge to make judgments about propositional similarities that determine safety.

    A proper evaluation of the knowledge-first program and the role of safety in it goes beyond the scope of this paper. (Cf. Bogardus 2014, note 1; Hirvelä 2019; Kelp 2017; Miracchi 2015; Williamson 2009a, for relevant discussions.) At any rate, the main target of this paper is the majority of safety theorists who intend the safety principle to be a non-circular, substantive condition for knowledge. At least for these theorists, my following critique regarding SAFETYSIMILAR applies.

  19. 19.

    An anonymous referee gives the following argument for determining propositional similarity in terms of subject matter of propositions: “In order to be properly safe one needs to be safe in one’s inquiry. This involves that one could not easily have falsely believed any propositions that belong to the subject matter of one’s inquiry. In other words, one could not easily have falsely believed any proposition that has the same subject matter as the proposition one actually believed”. In reply, I want to point out that this line of argument is susceptible to the kind of objection that I will raise against “Question-Relative Safety” in Sect. 5.4. To briefly anticipate, suppose that in Cynthia’s Perception, Cynthia has an explicit inquiry of figuring out what Jay is doing right now. With respect to this inquiry, she believes, and knows, that Jay is playing chess with another guy. That said, she could easily have falsely believed in a proposition that belongs to the same inquiry—the proposition that Jay is playing chess with Tim. Thus, the statement that one could not easily have believed any false propositions that belong to the subject matter of inquiry seems too strong.

  20. 20.

    See Littlejohn and Dutant (forthcoming) for an account of knowledge that also appeals to the notion of ‘normality’. Smith (2016) also defends a view according to which justification is understood in terms of ‘normality’.

  21. 21.

    Beddor and Pavese reject a statistical conception of normality. Nor do they think normal conditions are favorable conditions. They claim that “A more promising option is to identify the normal conditions for a task with those which we would consider to be fair for performing and assessing the task. We suspect it will be difficult to give a precise, non-circular analysis of what these conditions consist in. However, our intuitions about cases reveal a tacit grasp of these conditions. For example, in the case of shooting hoops, fair conditions include freedom from external intervention, sufficient lighting, etc. Moreover, it seems that we manifest our tacit conception of fair conditions when setting up competitions of various sorts” (Ibid. pp. 9–10).

  22. 22.

    Note here that we are supposing Beddor and Pavese take the first horn and thereby formulate safety along the line of SAFETYSAME.

  23. 23.

    I find it unclear whether or not Beddor and Pavese’s safety can judge Greco’s (2007) Frog correctly. If the relevant possible worlds for evaluating safety include the relatively remote ones where Jack sees non-green frogs in remote places—worlds in which he forms a false belief that he sees a green frog—then, plausibly, Jack’s belief would be counted as unsafe. However, in order to include those possible worlds, it has to be the case that conditions in those worlds are “at least as normal for the task at hand as in the actual world”. Given that Beddor and Pavase interpret “normal conditions” in terms of “fair conditions” (see note 21), it remains unclear whether the conditions in those possible worlds are as ‘fair’ as in the actual world or not. After all, what kind of conditions count as ‘fair’ for a color-blind person like Jack seems to be a largely underdetermined matter. Indeed, Beddor and Pavese (Ibid., note 23) admit that some of Greco’s (2007) cases are not susceptible to the normality-based safety they have offered. I think Frog is one of these cases.

  24. 24.

    Consider Broncano-Berrocal’s (2018) version of safety:

    I-Safety: If S knows that p via a method of belief formation M in appropriate determining conditions, then in nearly all (if not all) close possible worlds in which S continues to believe that p via M, the determining conditions for S’s belief that p continue to be appropriate (Broncano-Berrocal, ibid, p. 402).

    I-Safety, as it stands, is formulated in line with SAFETYSAME, since it focuses on error-possibilities of beliefs (only) in p. But I-Safety doesn’t seem to be able to avoid the first horn of the dilemma that afflicts SAFETYSAME. In particular, Frog still appears to be a counterexample to I-Safety. First, notice that in the actual world, the determining conditions for Jack’s forming a true belief that he sees a green frog are indeed appropriate. As we may assume, the lighting conditions are good, the frog is indeed green, etc. Besides, in all the nearby possible worlds where Jack sees a frog in the area, presumably these conditions continue to hold. Thus, Jack’s belief is safe according to I-Safety, which is implausible. (Compare my diagnosis here with Broncano-Berrocal’s diagnosis of the barn-façade case. See ibid., p. 403) Now, what if a defender of I-Safety takes the second horn of the dilemma, and thereby formulates safety in line with SAFETYDIFF? In that case, for a belief in p to be safe, the conditions should not only be appropriate relative to believing p, but also appropriate for other beliefs produced by the same method. And this would render I-Safety susceptible to Cynthia’s Perception: the conditions are inappropriate for Cynthia to form a belief that Jay is playing chess with Tim—after all, the conditions are indeed inadequate for Cynthia to reliably discriminate between Tim and his twin brother Tom. As a result, Cynthia’s @belief is unsafe, implausibly.

  25. 25.

    Peet and Pitcovski think that in most cases methods and explanations go hand in hand. See ibid, note 7.

  26. 26.

    Incidentally, notice the difference between “explanation of one’s holding/forming a belief” and “explanation of a belief being true”. The fact that Jack sees the frog does not explain why his belief is true, but it explains why he is holding/forming the belief.

  27. 27.

    See also Karjalainen and Morton (2003).

  28. 28.

    A defender of QRS needs to make a decision on whether QRS needs to be relativized to methods or not. I do not see any good reason not to do so. But this point need not bother us here.

  29. 29.

    I am assuming that in this scenario Cynthia’s @belief in the proposition that Jay is playing chess with another guy is formed upon answering this explicit question, just as in the following second scenario she forms the belief upon answering the second question.

  30. 30.

    Thanks to Blake Roeber for a helpful conversation.

  31. 31.

    Advocates of RVE include, among others, Broncano-Berrocal (2017), Carter (2016), Greco (2010, 2012), Navarro (2015, 2016), Sosa (2007, 2010, 2015), Turri (2011). See also Kelp (2017), Miracchi (2015) for ‘knowledge-first’ versions of RVE.

  32. 32.

    Most notably, see Pritchard (2012).

  33. 33.

    Greco (2012) presents an alternative reading of because relation in terms of pragmatic considerations. I have introduced Greco’s (2010) ‘explanatory salience’ account here because the latter is a better contrast to the target theory I will criticize in this section, namely Pritchard’s ALVE.

  34. 34.

    Another reason that Pritchard opts for a weaker ability condition is that he thinks one cannot deal with testimonial knowledge with the stronger ability condition. See his 2012.

  35. 35.

    Note that my critique below against ALVE also applies to Pritchard’s (2016) more recent anti-risk epistemology. At least, I do not see the latter has distinctive resources to resolve the problems against ALVE.

  36. 36.

    The point that Pritchard’s ability condition is satisfied in Mathema can be further strengthened by the fact that he thinks his (weak) ability condition is met in ordinary cases of testimonial knowledge (Pritchard 2012, pp. 273–274). If one’s acquiring truth by way of testimony can be the product of one’s relevant cognitive abilities, there seems no consistent reason to deny that one’s forming true beliefs from external equipment is also the product of relevant abilities. After all, relying on others and relying on equipment fall under the same general rubric of epistemic dependence.

  37. 37.

    Besides, presumably RVE can also handle Cynthia’s Perception. On Greco’s account, given that (1) Cynthia indeed perceives clearly that Jay is playing chess with another guy and (2) there are no abnormal factors involved with respect to Cynthia’s true belief in this proposition, her perceptual abilities can be plausibly counted as a salient factor in explaining her acquisition of this true belief. On Sosa’s account, since we have supposed that Cynthia has good perception and that she is sober, the constitution and the condition components are both satisfied. What about the situation component? Given that Tim indeed has a look-alike twin brother Tom, and that Cynthia cannot be expected to reliably discriminate between them in the current environment, it is plausible to think that the situation Cynthia is in is inappropriate for her to form a belief involving Tim. However, there is no reason to think that the situation here is inappropriate with respect to Cynthia’s @belief that Jay is playing chess with another guy. That is, given that there really is a guy playing chess with Jay and that Cynthia can be expected to reliably identify such a simple fact, the situation is appropriate for Cynthia to form @belief. And so, the situation component is satisfied with respect to @belief. Thus, @belief manifests Cynthia’s cognitive abilities, and this explains why she knows.

    An anonymous referee points out that the above remarks regarding Sosa pave way for a relatively more promising version of safety. Following Sosa, one can argue that a safe belief requires that the belief be formed via a ‘virtuous method’—method that is analyzed in terms of Sosa’s ‘constitution’, ‘condition/shape’, and ‘situation’. Now, an adequate evaluation of this proposal should include discussions of other recalcitrant cases in epistemology (e.g. barn façade case), which would require a separate paper. But I want to flag here that the proposal can give the following diagnosis of Cynthia’s Perception. The situation component is satisfied with regard to @belief. The component, however, is not satisfied in the nearby world where Cynthia forms a false belief involving Tim. Thus, the nearby world is irrelevant for determining whether @belief is safe (even if safety is formulated in terms of SAFETYDIFF), as the method is different in that world. (Cf. Greco 2016, Hirvelä 2019 who individuate methods in virtue-theoretic terms.).

    That said, the present proposal is still unpalatable for ALVE theorists. As Pritchard (2012) explains, ALVE is motivated by two “master intuitions” about knowledge: anti-luck intuition and ability(/virtue) intuition. He thinks that a safety condition and an independent ability/virtue condition capture these two intuitions, respectively. In particular, he claims that one can satisfy the safety condition without relevant cognitive abilities: “…it is to be expected that one could satisfy such a condition [safety] while not exhibiting any cognitive ability, since whatever modal requirement is imposed, with imagination one could think of a way in which it can be satisfied in a manner that bears no relation to the agent’s cognitive abilities.” (Ibid., p. 272) This is clearly inconsistent with the present proposal—the latter does require that a safe belief should be formed via relevant cognitive abilities.

  38. 38.

    Greco (2016), defends the following safety condition:

    (Ability + Proper Conditions)-relative Safety: A belief is (Ability + Proper Conditions)-relative safe just in case: In close worlds where S believes p from ability A, and in conditions proper for the exercise of A, p is true (p. 54).

    Greco also takes Sosa (2007) to be committed to such a safety principle. In fact, both Greco and Sosa interpret cognitive abilities in a modal way. Cognitive abilities are not about success rates in the actual world; rather, they are taken to be reliable dispositions to achieve truths in relevant counterfactual situations. Therefore, whenever one knows, and for that matter one believes truth because of cognitive abilities, the abilities in question allow the subject to achieve truths in the majority of relevant counterfactual situations, thus satisfying the above safety condition (See Sosa 2007, p. 29; Greco 2016, pp. 54–55).

    Now, if RVE theorists like Greco and Sosa are committed to the above safety condition, one may think that the dilemma against safety also poses a threat to their views. However, I disagree. Note that Sosa (2007) and Greco (2016) do not advocate safety as an independently necessary condition for knowledge. Instead, safety is merely entailed by the ability condition, without safety itself doing any theoretical work over and above the ability condition that entails it. This means that, unlike ALVE theorists like Pritchard, Greco and Sosa can appeal to their ability conditions to account for Mathema, without worrying if such a move endangers the motivation for their views. That is because, for them, the safety condition is not supposed to play any theoretical role (such as eliminating epistemic luck) over and above the ability condition itself. (I thank John Greco for bringing more clarity to me on this issue.)


  1. Alspector-Kelly, M. (2011). Why safety doesn’t save closure. Synthese, 183(2), 127–142.

    Article  Google Scholar 

  2. Ball, B. (2016). Knowledge, safety, and questions. Filosofia Unisinos, 17(1), 58–62.

    Article  Google Scholar 

  3. Baumann, P. (2008). Is knowledge safe? American Philosophical Quarterly, 45, 19–30.

    Google Scholar 

  4. Beddor, B., & Pavese, C. (2018). Modal virtue epistemology. Philosophy and Phenomenological Research.

    Article  Google Scholar 

  5. Bernecker, S. (2010). Memory: A philosophical study. Oxford: Oxford University Press.

    Google Scholar 

  6. Bernecker, S. (forthcoming). Against global method safety. Synthese

  7. Bishop, M. (2010). Why the generality problem is everybody’s problem. Philosophical Studies, 151(2), 285–298.

    Article  Google Scholar 

  8. Bogardus, T. (2014). Knowledge under threat. Philosophy and Phenomenological Research, 86(1), 289–313.

    Article  Google Scholar 

  9. Bogardus, T., & Marxen, C. (2014). Yes, safety is in danger. Philosophia, 42(2), 321–334.

    Article  Google Scholar 

  10. Broncano-Berrocal, F. (2014). Is safety in danger? Philosophia, 42(1), 63–81.

    Article  Google Scholar 

  11. Broncano-Berrocal, F. (2017). A Robust Enough Virtue Epistemology. Synthese, 194(6), 2147–2174.

    Google Scholar 

  12. Broncano-Berrocal, F. (2018). Knowledge and tracking revisited. Analysis, 78(3), 396–405.

    Article  Google Scholar 

  13. Carter, J. A. (2010). Anti-luck epistemology and safety’s discontents. Philosophia, 38(3), 517–532.

    Article  Google Scholar 

  14. Carter, J. A. (2016). Robust virtue epistemology as anti-luck epistemology: A new solution. Pacific Philosophical Quarterly, 97(1), 140–155.

    Article  Google Scholar 

  15. Comesaña, J. (2005). Unsafe knowledge. Synthese, 146(3), 395–404.

    Article  Google Scholar 

  16. Freitag, W. (2014). Safety, sensitivity and “distant” epistemic luck. Theoria, 80(1), 44–61.

    Article  Google Scholar 

  17. Goldman, A. (1976). Discrimination and perceptual knowledge. Journal of Philosophy, 73(November), 771–791.

    Article  Google Scholar 

  18. Goodman, N. (1972). Seven strictures on similarity. In N. Goodman (Ed.), Problems and projects (pp. 437–447). New York: BobbsMerrill.

    Google Scholar 

  19. Greco, J. (2007). Worries about Pritchard’s safety. Synthese, 158, 299–302.

    Article  Google Scholar 

  20. Greco, J. (2010). Achieving knowledge: A virtue-theoretic account of epistemic normativity. Cambridge: Cambridge University Press.

    Book  Google Scholar 

  21. Greco, J. (2012). A (different) virtue epistemology. Philosophy and Phenomenological Research, 85(1), 1–26.

    Article  Google Scholar 

  22. Greco, J. (2016). Knowledge, virtue and safety. In Performance Epistemology (Ed.), Miguel Ángel Fernández (pp. 51–61). New York: Oxford University Press.

    Google Scholar 

  23. Grundmann, T. (forthcoming). Saving safety from counterexamples. Synthese.

  24. Hiller, A., & Neta, R. (2007). Safety and epistemic luck. Synthese, 158(3), 303–313.

    Article  Google Scholar 

  25. Hirvelä, J. (2018). No safe Haven for the virtuous. Episteme.

    Article  Google Scholar 

  26. Hirvelä, J. (2019). Global safety: How to deal with necessary truths. Synthese, 196(3), 1167–1186.

    Article  Google Scholar 

  27. Karjalainen, A., & Morton, A. (2003). Contrastive knowledge. Philosophical Explorations, 6(2), 74–89.

    Article  Google Scholar 

  28. Kelp, C. (2009). Knowledge and safety. Journal of Philosophical Research, 34, 21–31.

    Article  Google Scholar 

  29. Kelp, C. (2017). Knowledge first virtue epistemology. In Emma Gordon, Benjamin Jarvis, & Adam Carter (Eds.), Approaches in epistemology and mind (pp. 223–245). Knowledge First: Oxford University Press.

    Google Scholar 

  30. Kripke, S. (2011). Nozick on knowledge. Philosophical troubles: Collected papers (Vol. 1, pp. 162–224). Oxford: Oxford University Press.

    Google Scholar 

  31. Kvanvig, J. (2008). Epistemic luck by Duncan Pritchard. Philosophy and Phenomenological Research, 77, 272–281.

    Article  Google Scholar 

  32. Lasonen-Aarnio, M. (2010). Unreasonable knowledge. Philosophical Perspectives, 24(1), 1–21.

    Article  Google Scholar 

  33. Littlejohn, C. & Dutant, J. (forthcoming). Justification, knowledge, and normality. Philosophical Studies.

  34. Manley, D. (2007). Safety, content, apriority, self-knowledge. Journal of Philosophy, 104(8), 403–423.

    Article  Google Scholar 

  35. McEvoy, M. (2014). Causal tracking reliabilism and the gettier problem. Synthese, 191(17), 4115–4130.

    Article  Google Scholar 

  36. Miracchi, L. (2015). Competence to know. Philosophical Studies, 172(1), 29–56.

    Article  Google Scholar 

  37. Navarro, J. (2015). No achievement beyond intention. Synthese, 192(10), 3339–3369.

    Article  Google Scholar 

  38. Navarro, J. (2016). Acting in order to know, knowing in order to act: Sosa on epistemic and practical deliberation. Disputatio, 8(43), 233–252.

    Article  Google Scholar 

  39. Neta, R., & Rohrbaugh, G. (2004). Luminosity and the safety of knowledge. Pacific Philosophical Quarterly, 85(4), 396–406.

    Article  Google Scholar 

  40. Nozick, R. (1981). Philosophical explanations. Oxford: Oxford University Press.

    Google Scholar 

  41. Peet, A., & Pitcovski, E. (2018). Normal knowledge: Toward an explanation-based theory of knowledge. Journal of Philosophy, 115(3), 141–157.

    Article  Google Scholar 

  42. Pritchard, D. (2005). Epistemic luck. Oxford: Oxford University Press.

    Book  Google Scholar 

  43. Pritchard, D. (2007). Anti-luck epistemology. Synthese, 158(3), 277–297.

    Article  Google Scholar 

  44. Pritchard, D. (2012). Anti-luck virtue epistemology. Journal of Philosophy, 109(3), 247–279.

    Article  Google Scholar 

  45. Pritchard, D. (2015). Anti-luck epistemology and the Gettier problem. Philosophical Studies, 172(1), 93–111.

    Article  Google Scholar 

  46. Pritchard, D. (2016). Epistemic risk. Journal of Philosophy, 113(11), 550–571.

    Article  Google Scholar 

  47. Rabinowitz, D. (2011). The safety condition for knowledge. Internet Encyclopedia of Philosophy.

  48. Sainsbury, R. M. (1997). Easy possibilities. Philosophy and Phenomenological Research, 57(4), 907–919.

    Article  Google Scholar 

  49. Schaffer, J. (2005). Contrastive knowledge. In T. S. Gendler & J. Hawthorne (eds.), Oxford studies in epistemology (Vol. 1, pp. 235–271). Oxford University Press.

  50. Schaffer, J. (2007). Knowing the answer. Philosophy and Phenomenological Research, 75(2), 383–403.

    Article  Google Scholar 

  51. Schaffer, J. (2008). Knowledge in the image of assertion. Philosophical Issues, 18(1), 1–19.

    Article  Google Scholar 

  52. Smith, M. (2016). Between probability and certainty: What justifies belief. Oxford: Oxford University Press.

    Book  Google Scholar 

  53. Sosa, E. (1999). How to defeat opposition to Moore. In J. Tomberlin (Ed.), Philosophical perspectives 13: Epistemology (pp. 141–154). Oxford: Blackwell.

    Google Scholar 

  54. Sosa, E. (2004). Relevant alternatives, contextualism included. Philosophical Studies, 119(1–2), 35–65.

    Article  Google Scholar 

  55. Sosa, E. (2007). A virtue epistemology: Apt belief and reflective knowledge (Vol. I). Oxford: Oxford University Press.

    Book  Google Scholar 

  56. Sosa, E. (2010). How competence matters in epistemology. Philosophical Perspectives, 24(1), 465–475.

    Article  Google Scholar 

  57. Sosa, E. (2015). Judgment & agency. Oxford: Oxford University Press.

    Book  Google Scholar 

  58. Turri, J. (2011). Manifest failure: The Gettier problem solved. Philosophers’ Imprint, 11(8), 1–11.

    Google Scholar 

  59. Vogel, J. (2007). Subjunctivitis. Philosophical Studies, 134(1), 73–88.

    Article  Google Scholar 

  60. Weatherson, B. (2004). Luminous margins. Australasian Journal of Philosophy, 82(3), 373–383.

    Article  Google Scholar 

  61. Whiting, D. (forthcoming). Knowledge, justification, and (a sort of) safe belief. Synthese.

  62. Williamson, T. (2000). Knowledge and its limits. Oxford: Oxford University Press.

    Google Scholar 

  63. Williamson, T. (2009a). Reply to Goldman. In Duncan Pritchard & Patrick Greenough (Eds.), Williamson on knowledge (pp. 305–312). Oxford: Oxford University Press.

    Google Scholar 

  64. Williamson, T. (2009b). Reply to John Hawthorne and Maria Lasonen-Aarnio. In Duncan Pritchard & Patrick Greenough (Eds.), Williamson on knowledge (pp. 313–329). Oxford: Oxford University Press.

    Google Scholar 

  65. Zhao, H. (forthcoming). Knowledge without safety. Synthese.

Download references


Earlier drafts of this paper were presented at University of Notre Dame (Fall 2018), Saint Louis University (Spring 2019), and Pacific APA (April 2019). Thanks to the audiences at these occasions. Special thanks to Robert Audi, Blake Roeber and Yingjin Xu for helpful discussions. I am also very grateful to Peter Baumann, Julianne Chung, John Greco, Amy Milton, Joe Salerno, Yiling Zhou, and two anonymous referees of this journal for much helpful feedback on previous drafts.

Author information



Corresponding author

Correspondence to Haicheng Zhao.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Zhao, H. Better virtuous than safe. Synthese 198, 6969–6991 (2021).

Download citation


  • Safety
  • Necessary proposition
  • Virtue epistemology
  • Robust virtue epistemology
  • Anti-luck virtue epistemology